Research Article Reversible Data Hiding Scheme with High ...SMD ( , cw ) = (S (X ,U ,L ),B (cw )) =...
Transcript of Research Article Reversible Data Hiding Scheme with High ...SMD ( , cw ) = (S (X ,U ,L ),B (cw )) =...
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2013 Article ID 476181 11 pageshttpdxdoiorg1011552013476181
Research ArticleReversible Data Hiding Scheme with High Embedding CapacityUsing Semi-Indicator-Free Strategy
Jiann-Der Lee1 Yaw-Hwang Chiou1 and Jing-Ming Guo2
1 Department of Electrical Engineering Chang Gung University Tao-Yuan 333 Taiwan2Department of Electrical Engineering National Taiwan University of Science and Technology Taipei 106 Taiwan
Correspondence should be addressed to Jing-Ming Guo jmguoseednettw
Received 12 June 2013 Accepted 5 September 2013
Academic Editor Marco Perez-Cisneros
Copyright copy 2013 Jiann-Der Lee et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
A novel reversible data-hiding scheme is proposed to embed secret data into a side-matched-vector-quantization- (SMVQ-)compressed image and achieve lossless reconstruction of a vector-quantization- (VQ-) compressed image The rather randomdistributed histogramof aVQ-compressed image can be relocated to locations close to zero by SMVQpredictionWith this strategyfewer bits can be utilized to encode SMVQ indices with very small valuesMoreover no indicator is required to encode these indiceswhich yields extrahiding space to hide secret data Hence high embedding capacity and low bit rate scenarios are deposited Morespecifically in terms of the embedding rate the bit rate and the embedding capacity experimental results show that the performanceof the proposed scheme is superior to those of the former data hiding schemes for VQ-based VQSMVQ-based and search-order-coding- (SOC-) based compressed images
1 Introduction
With the explosive growth of communication through theInternet information processing and management at anytime have become a standard service for most people sub-sequently inducing security problems such as interceptionmodification and montage Thus one critical issue is howto find effective ways to protect information transmissionover the Internet with safety and security To cope with thisdata hiding techniques have attracted attentions for beingable to embed secret data into a cover image with minimalperceptual degradation
In general data hiding techniques can be classifiedinto two categories namely reversible data hiding schemesand irreversible data hiding schemes For irreversible datahiding schemes only secret data can be extracted whilerestoration of cover images is unavailable The irreversibledata hiding scheme is not suited for certain applications thatdemand the original cover image to be unaltered by the datahiding process Conversely reversible data hiding schemescan extract the secret data and recover the original coverimages in the decoder Moreover data hiding schemes canbe performed in three possible formats that is in spatial
domain [1ndash5] in frequency domain [6ndash8] and in compresseddomain [9ndash15] In spatial domain Tian [1] proposed ascheme for reversible data hiding called difference expansion(DE) Tianrsquos work expands the value differences between twoneighboring pixels to embed a bit This scheme is easy torealize yet the embedding capacity heavily depends on thesmoothness of an image Ni et al [2] utilized the zero of theminimum points of the histogram in an image and slightlymodified the pixel values to embed secret data into the peakpoints of the histogram of an image The scheme is simplefor implementation and which supports high visual qualityHowever the embedding capacity is restricted by the peakpoints of the histogram of an image Moreover a cover imagecan be transformed into the frequency domain with variouspossible transform kernels for example DCT DFT or DWTNormally frequency domain techniques embed message bymodulating the transformed coefficients of subbands withjust noticeable difference (JND) according to the sensitivityof the human visual system (HVS)
Data hiding techniques in compressed domain can relishthe both advantages of data hiding and compression for amultimedia distribution Jo and Kim [9] firstly proposed anirreversible VQ-based scheme which is easy to realize yet
2 Mathematical Problems in Engineering
the hiding capacity is rather small Later many VQSMVQ-based schemes with indicators are proposed For exampleChang et al [10] developed a VQ-based data hiding withrecovery capability The scheme has low embedding capacityand high bit rate (BR) because of an indicator added in frontof most encoded indices and only two clusters are used tohide secret data In [11 12] a secret data hiding was proposedbased on the search-order coding compression method ofVQ indices to increase embedding capacityThe search-ordercoding selects the neighboring indices of the encoding indexto form a state codebook and then search-order code (SOC)and original index value (OIV) are employed to hide a secretbit 0 or 1 A 1-bit indicator is always employed to distinguishbetween SOC and OIV For this scheme the original coverimage can be restored completely Shie et al [13] developedan adaptive data hiding based on VQ-compressed imagein which image blocks are classified into embeddable andunembeddable blocks based on the variances and side-matchdistortions (SMDs) The embeddable blocks are employed tohide secret data while those unembeddable blocks remainunchanged It is necessary that indicator be used for the blockjudgment Although this scheme can yield a high embeddingcapacity the quality of the reconstructed image reduces as thequantity of the secret data increases In addition this schemeis a case of irreversible information hiding In [14] both VQand SMVQ are used to hide secret data in which a 1-bitindictor is always required for each index and only 1-bit secretdata can be embedded in each index when the VQ techniqueis applied Conversely more than one bit secret data canbe embedded in each index when the SMVQ technique isapplied This scheme can provide a higher hiding capacityyet it is an irreversible data-hiding scheme In [15] Yang andLin extended the scheme proposed by Chang et al [10] bydividing the VQ codebook into 2BS clusters and half of whichare used to embed secret data where BS denotes the size ofthe secret data embedded into each VQ index In the schemeboth of the VQ and SMVQ are applied to hide secret data anda 1-bit indictor is required for the index identification More-over it is a reversible data-hiding scheme In these schemesindicators are always used which causes a higher bit rate
By exploiting the special indices distribution of a SMVQ-compressed image that frequently used indices are encodedby short codes while rarely-used indices are encoded by longcodes this work proposed a novel reversible data hiding inthe SMVQ-compressed domain The scheme can increasethe hiding capacity and completely restore the cover imagefrom an embedded image after the hiding data is extractedMoreover the indices of a VQ-compressed image can berelocated close to zero by using SMVQ predications SMVQindices located around zero can be encoded using fewer bitsCompared with the previous schemes [10 12 15] in whichthe bit rates are increased because indicators are requiredin most of the indices the proposed scheme employs theSMVQ predication and thus most of the indices are encodedwithout indicator As a result high embedding capacity andlow bit rate scenarios can be achieved As documented inthe experimental results the performance of the proposedscheme is significantly superior to that of the previous works
[10 12 15] in terms of the embedding rate bit rate andembedding capacity
The rest of this paper is organized as follows Section 2presents the related works to briefly introduce the conceptsof VQ and SMVQ Section 3 describes the proposed schemein detail Section 4 gives the experimental results and per-formances comparisons of the proposed scheme with formerapproaches Finally Section 5 draws conclusions
2 Backgrounds
21 Brief Concept of the VQ VQ initially involves codebookconstruction from a set of training images the trainingimages are partitioned into nonoverlapping blocks and themost representative blocks are selected to form codebook inwhich the elements in the codebook are called codewordsIn general the LBG algorithm [16] is employed to producethe desired codebook With the generated codebook eachblock in an image is encoded with the index of the nearestcodeword such that the total storage space for an imageis minimized To measure the similarity between a block119861 = (119887
1 1198872 119887
119896) and a codeword 119862
119895= (1198881198951 1198881198952 119888
119895119896)
where 119862119895is the 119895th codeword in the codebook the squared
Euclidean distance is generally used as below
119863(119861 119862119895) =
119896
sum
119894=1
(119887119894minus 119888119895119894)2
(1)
When the encoding process is completed block 119861 issimply represented by the index min 119895 where 119862min 119895 is thenearest codeword of the block 119861 For decoding process thetable lookup is performed with the same codebook as thatused by the encoder
22 SMVQ Kim initially proposed SMVQ for image coding[17] SMVQ exploits the high correlation among neighboringblocks and side-match prediction to create a state codebookwith a size smaller than that of the original VQ codebookThe side-match prediction assumes that the values of theadjacent pixels between the neighboring blocks are equal InSMVQ blocks located in the first row and first column ofan image are encoded by traditional VQ and the remainingblocks are predicted using the corresponding state codebookFigure 1 shows an example of the relationships among anencoding block119883 its upper neighboring block 119880 and its leftneighboring block 119871 for blocks of size 4times4This work denotesthe border vector and the side vector of block 119883 as 119861(119883) =
(11988311198832119883311988341198835119883911988313) = (1198871 1198872 1198873 1198874 1198875 1198876 1198877) and
119878(119883119880 119871) = ((11988013 + 1198714)2 11988014 11988015 11988016 1198718 11987112 11987116) =
(1199041 11990421199043 1199044 1199045 1199046 1199047) The SMD between a block 119883predicted
by blocks 119880 and 119871 and a codeword cw in the codebookis measured by the squared Euclidean distance denoted asfollows
SMD (119883 cw) = 119863 (S (XU L) B (cw)) =7
sum
119894=1
(119904119894minus 119887119894)2
(2)
For each block SMVQ sorts the codebook CB accordingto SMDs and thus the first 119865
119888codewords in the sorted
Mathematical Problems in Engineering 3
U13 U14 U15 U16L4 X1 X2 X3 X4L8 X5L12 X9L16 X13
Figure 1 An example of SMVQ
codebook are picked to form the state codebook SC in whichSMDs are the smallest for block 119883 Subsequently the SMVQsearches the closest codeword from the state codebook forblock 119883 and then the index of the closest codeword is usedto encode block 119883 Because the size of the state codebook issmaller than that of the VQ codebook the size of the SMVQindex is reduced to improve the coding gain However thereis a distortion between the block decoded by the SMVQindex and the corresponding block decoded by the VQindex
3 The Proposed Scheme
A grayscale cover image 119862 of size 119873 lowast 119872 is partitionedinto nonoverlapped blocks BLKi119894=1119871 of 119908 lowast ℎpixels inwhich 119871 = (119873119908) lowast (119872ℎ) Normally the variables 119908 =
ℎ = 4 Each block of the image can be encoded using VQand thus each block is represented with an index of thenearest codeword in the codebook AVQ-compressed imagedenoted by 119868 consists of indices The size of an index isproportional to the number of codewords in the codebookA SMVQ-compressed image is generated by applying SMVQpredictions to a VQ-compressed image To reconstruct a VQ-compressed image completely the size of the state codebookand that of the VQ codebook are first set as identical Nextthe codebook is sorted according to the SMDs which denotethe differences between an encoding block and the codewordsin the VQ codebook Finally the state codebook is generatedusing all of the codewords in the sorted codebook For a VQindex the corresponding SMVQ index can be obtained usingthe state codebook and no distortion is presented between thetwo codewords decoded by the two indices
Figures 2 and 3 show histograms of the VQ-compressedPeppers image and the corresponding SMVQ-compressedimage respectively with a codebook of size 256 Basedon the SMVQ predication most indices of the SMVQ-compressed image are distributed close to zero For a SMVQ-compressed image a very small value 119873
119911binary digits (119873
119911
bits) representing unsigned integers from 0 to 2119873119911
minus 1 cancover most of the indices In the proposed scheme the 2119873119911numbers are employed as indicators orand encoding codesTable 1 shows the functions of each number of the119873
119911binary
digits (119873119911bits) A binary number among 0 and 2
119873119911minus 3
represents an encoding code and indicator Binary numbers
0
100
200
300
400
500
600
700
800
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
Figure 2 Histogram of the VQ-compressed Peppers image
0
1000
2000
3000
4000
5000
6000
7000
8000
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
9000
Figure 3 Histogram of the SMVQ-compressed Peppers image
Table 1 The functions of each number of the 119873119911binary digits
(119873119911bits)
Number Function0sim2119873119911 minus 3 Indicator and encoding code2119873119911 minus 2 Indicator (indicator 0)
2119873119911 minus 1 Indicator (indicator 1)
2119873119911
minus2 and 2119873119911 minus1 represent indicators denoted by indicator0 and indicator 1 respectively
31 The Encoding Process Suppose the size of a secret datasd is 119904 bits During the encoding process a SMVQ indexwith a value less than 2
119873119911minus 2 which is defined as a tiny-
value index is denoted by119864119894 and the corresponding encoded
index is denoted by 1198641015840
119894 Then 1198641015840
119894is expressed in (3) where
and and denote exponential and concatenation operatorsrespectively
1198641015840
119894= 119864119894 sd (3)
4 Mathematical Problems in Engineering
Table 2 The ranges and symbols of tiny-value middle-value andlarge-value indices
Indices Range SymbolTiny-value indices 0sim2119873119911 minus 3 119864
119894
Middle-value indices 0 + 2119873119911minus 2sim2119873119888 minus 1 + 2
119873119911 minus 2 119866
119894
Large-value indices 2119873119888 + 2119873119911minus 2simcsminus 1 119874
119894
Most of the SMVQ indices can be encoded using (3)including no indicator and the rest are encoded with the helpof indicator 0 or indicator 1 However the adoption of theindicator increases the size of the encoded result A SMVQindex with a value in between (2
119873119911minus 2) + 0 and (2119873119911 minus 2) +
2119873119888minus1 is defined as amiddle-value index Indices withmiddle
values are encoded using indicator 0 concatenated with anencoding space with119873
119888bits Amiddle-value index is denoted
by119866119894 and the corresponding encoded index is denoted by1198661015840
119894
which can be expressed in the following
1198661015840
119894= 119894119899119889119894119888119886119905119900119903 0 119866
119894minus 2119873119911
+ 2 (4)
The119873119888-bit encoding space of (4) is not to hide data but to
collect more indices Indicator 1 is employed to assist encod-ing a SMVQ index which cannot be encoded using (3) or (4)A SMVQ index with a value more than (2119873119911 minus 2) + 2
119873119888minus 1
is defined as a large-value index denoted by 119874119894 and the
corresponding encoded index is denoted by 1198741015840119894which can be
expressed in the following
1198741015840
119894= 119894119899119889119894119888119886119905119900119903 1 119874
119894 (5)
Equation (5) not only hides no data but also increases anadditional indicator with 119873
119911bits For a SMVQ-compressed
image most indices with tiny values are encoded using (3)which hides 119904 bits and generates an encoded result with119873
119911+119904
bits for each index a few indices are encoded using (4)which generates an encoded result with119873
119911+119873119888bits for each
index few indices are encoded using (5) which generates anencoded result with119873
119911+log2(cs) bits for each index where cs
indicates the size of a codebook Table 2 shows the ranges andsymbols of tiny-value middle-value and large-value indicesThe size of an embedded image can be measured with (6)where 119909 denotes the size of 119909 and119868
119911 119868119888 and 119868
119900denote the
number of indices with tiny values middle values or largevalues respectively
1003817100381710038171003817embedded image1003817100381710038171003817 = (119873119911+ 119904) lowast 119868
119911
+ (119873119911+ 119873119888) lowast 119868119888
+ (119873119911+ log2(cs)) lowast 119868
119900
(6)
In (6) (119873119911+ 119904) and (119873
119911+ 119873119888) can be controlled in less
or more than log2(cs) bits and only the third item always
requires a space with more than log2(cs) bits It is remarkable
that a larger 119904 leads to a larger embedding capacity and viceversa Moreover a larger 119873
119888increases 119868
119888 while decreasing
119868119900 and vice versa Based on the special index distribution
property of the SMVQmost indices can be encodedusing (3)
An encoding index with value Q
Is Q in between(2Nz
minus 2) + 0 and (2Nzminus 2) + 2
Ncminus 1
A middle-valueindex encodedusing (4)
A large-valueindex encodedusing (5)
A tiny-valueindex encodedusing (3)
No
No
Yes
YesQ le 2
N119911 minus 3
Figure 4 An example of the index classification and encodingschemes
in which no indicator is required and 119904-bit data are hiddenin each index and thus a high embedding capacity and alow embedded image size can be achieved in the encodingprocess The algorithm of the embedding process is summa-rized as follows and an example of the index classification andencoding schemes is shown in Figure 4
Input A grayscale cover image 119862 of size119873 lowast119872 a codebookCB of size cs a secret data stream and parameters119873
119911and119873
119888
OutputThe code stream in binary form cs 119887 and parameters119873119911and119873
119888
Step 1 Compress the cover image 119862 using VQ to obtain theVQ-compressed image 119868 of size (119873119908) lowast (119872ℎ)
Step 2 Perform SMVQ predictions with the VQ-compressedimage 119868 to create the SMVQ-compressed image 119879 of size(119873119908) lowast (119872ℎ)
Step 3 Read the next SMVQ index denoted as 119876 from theSMVQ-compressed image 119879 with the raster scan order
Step 4 If 119876 le 2119873119911 minus 3 (a tiny-value index) then we have the
following
Step 41 Read 119904-bit data sd from the secret data stream
Step 42 119876 is encoded using (3) cs 119887 = cs 119887 D2B(119876) sdwhere denotes a concatenation operation and D2B denotesa decimal to binary operation
Step 5 If119876 is in between (2119873119911 minus 2) + 0 and (2119873119911 minus 2) + 2
119873119888 minus 1
(a middle-value index) then 119876 is encoded using (4) cs119887=
cs 119887 119894119899119889119894119888119886119905119900119903 0 D2B(119876 minus 2119873119911 + 2)
Mathematical Problems in Engineering 5
(a)
00000 01100 00110 01010 00010 01111 001100 12 6 10 2 15 6
(b) (c)
0 1 2 3
4 5 1 7
8 0 1 10
127 2 1 0 1 || 15
1 || 2
1 || 6
2 || 0
3 || 127
2 || 6
2 || 2
0 || 0
2 || 0
0 || 10
2 || 3
1 || 12
0 || 6
2 || 8
2 || 5
2 || 1
Figure 5 An example of the encoding process for hiding 5-bit secretdata (a) Secret data (b) SMVQ-compressed image and (c) encodedimage
Step 6
If 119876 ge (2119873119911 minus 2) + 2
119873119888 (a large-value index) then
119876 is encoded using (5)cs 119887 = cs 119887 119894119899119889119894119888119886119905119900119903 1 D2B(119876)
Step 7 Repeat Steps 3ndash7 until all of the indices in the SMVQ-compressed image are processed
Step 8 Output the code stream cs 119887 and parameters 119873119911and
119873119888
Figure 5 shows an example of the encoding process forhiding 5-bit secret data Let 119868
0be indicator 0 and 119868
1indicator
1 where 119873119911= 2 119873
119888= 5 119904 = 5 and 119888119904 = 128 and the size
of the state codebook is identical to that of the VQ codebookFor each image (Figure 5) let 119877
1to 1198774be the indices located
in the first row and 1198775to 1198778the indices located in the second
row and so on For the SMVQ-compressed image 1198771has a
value 0 and it is less than 2119873119911 minus2 The index is encoded using(3) Secret data are (00000)
2 According to (3) the index with
119873119911bits followed by secret data (00000)
2is presented as the
encoded result for 1198771 1198774has a value of 3 and it is in between
(2119873119911
minus 2) + 0 and (2119873119911
minus 2) + 2119873119888
minus 1 An indicator 1198680 2
followed by the offset index 1 is the encoded result for119877411987713
has a value of 127 and it is more than (2119873119911
minus 2) + 2119873119888
minus 1 Anindicator 119868
1 3 followed by the index 127 is the encoded result
for 11987713 All of the encoded results are in binary form and the
length of the encoded result of each index is (119873119911+119904) (119873
119911+119873119888)
or (119873119911+ log2(cs)) bits using (3) (4) or (5) respectively
32TheExtraction andReversion Process Thegoal of the dataextraction and reversion process is to extract the embeddedsecret data from an embedded image and then recover theoriginal VQ-compressed imageThe extraction and reversionprocess is introduced as below For an encoded index 119873
119911
bits are firstly read from code stream cs 119887 and the 119873119911-bit
binary value is converted to the decimal value denoted by119889 119873119911 In the first case if 119889 119873
119911is less than 2
119873119911minus 2 the
encoded index is an 119864119894index Then the original index is
directly restored by 119889 119873119911 Next read 119904 bits from the code
stream 119888 119887119904 to reconstruct the secret data In the second case
if 119889 119873119911is 2119873119911 minus 1 the encoded index is an 119874
119894index Then
read the next log2(cs) bits and convert them into decimal
value denoted by 119889 119874119894 the original index is recovered by
119889 119874119894 In the third case if 119889 119873
119911is 2119873119911 minus 2 the encoded index
is a 119866119894index Then read the next 119873
119888bits and convert them
into decimal value 119889 119866119894 the original index is recovered by
119889119866119894
+ (2119873119885 minus 2) When the original SMVQ-compressed image
has been reconstructed SMVQ predictions can be employedto restore the original VQ-compressed image In summarythe extraction and reversion process of the proposed schemeis organized as below
Input The code stream in binary form cs 119887 and parameters119873119911and119873
119888
Output The reconstructed VQ-compressed image 1198681015840 of size119873 lowast119872 and the retrieved secret data stream
Step 1 Set ptr = 1 and the restored sd = Null
Step 2 Read the next119873119911bits from the code stream 119888 119887119904 and
convert the119873119911bits to decimal value denoted by119889 119873
119911119889 119873119911=
B2D(cs 119887(ptr ptr + 119873119911minus 1) ptr = ptr + 119873
119911
Step 3 If 119889 119873119911lt 2119873119911
minus 2 (a tiny-value index) then
restored index = 119889 119873119911
read the next 119904 bits denoted by sd1015840 from code stream119888 119887119904restored sd = restored sd sd1015840ptr = ptr + 119904
Step 4 If 119889 119873119911= 2119873119911
minus 1 (indicator 1) then
read the next log2(cs) bits from the code stream c bs
119889 119874119894= B2D(cs 119887(ptr ptr + log
2(cs) minus 1) ptr = ptr +
log2(cs)
restored index = 119889 119874119894
Step 5 If 119889 119873119911= 2119873119911
minus 2 (indicator 0) then
read the next119873119888bits from the code stream 119888 119887119904
119889 119866119894= B2D(cs 119887(ptr ptr + 119873
119888minus 1) ptr = ptr + 119873
119888
restored index = 119889 119866119894+ (2119873119911
minus 2)
Step 6 Repeat Steps 2ndash6 until all of the bits in the code streamare processed
Step 7 Apply the SMVQ prediction to the reconstructedSMVQ-compressed image to obtain the original VQ-compressed image
Step 8 Output the reconstructed VQ-compressed image andthe restored secret data stream
4 Experimental Results
In this section we present the experimental results forevaluating the performance of the proposed scheme in terms
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Mathematical Problems in Engineering
the hiding capacity is rather small Later many VQSMVQ-based schemes with indicators are proposed For exampleChang et al [10] developed a VQ-based data hiding withrecovery capability The scheme has low embedding capacityand high bit rate (BR) because of an indicator added in frontof most encoded indices and only two clusters are used tohide secret data In [11 12] a secret data hiding was proposedbased on the search-order coding compression method ofVQ indices to increase embedding capacityThe search-ordercoding selects the neighboring indices of the encoding indexto form a state codebook and then search-order code (SOC)and original index value (OIV) are employed to hide a secretbit 0 or 1 A 1-bit indicator is always employed to distinguishbetween SOC and OIV For this scheme the original coverimage can be restored completely Shie et al [13] developedan adaptive data hiding based on VQ-compressed imagein which image blocks are classified into embeddable andunembeddable blocks based on the variances and side-matchdistortions (SMDs) The embeddable blocks are employed tohide secret data while those unembeddable blocks remainunchanged It is necessary that indicator be used for the blockjudgment Although this scheme can yield a high embeddingcapacity the quality of the reconstructed image reduces as thequantity of the secret data increases In addition this schemeis a case of irreversible information hiding In [14] both VQand SMVQ are used to hide secret data in which a 1-bitindictor is always required for each index and only 1-bit secretdata can be embedded in each index when the VQ techniqueis applied Conversely more than one bit secret data canbe embedded in each index when the SMVQ technique isapplied This scheme can provide a higher hiding capacityyet it is an irreversible data-hiding scheme In [15] Yang andLin extended the scheme proposed by Chang et al [10] bydividing the VQ codebook into 2BS clusters and half of whichare used to embed secret data where BS denotes the size ofthe secret data embedded into each VQ index In the schemeboth of the VQ and SMVQ are applied to hide secret data anda 1-bit indictor is required for the index identification More-over it is a reversible data-hiding scheme In these schemesindicators are always used which causes a higher bit rate
By exploiting the special indices distribution of a SMVQ-compressed image that frequently used indices are encodedby short codes while rarely-used indices are encoded by longcodes this work proposed a novel reversible data hiding inthe SMVQ-compressed domain The scheme can increasethe hiding capacity and completely restore the cover imagefrom an embedded image after the hiding data is extractedMoreover the indices of a VQ-compressed image can berelocated close to zero by using SMVQ predications SMVQindices located around zero can be encoded using fewer bitsCompared with the previous schemes [10 12 15] in whichthe bit rates are increased because indicators are requiredin most of the indices the proposed scheme employs theSMVQ predication and thus most of the indices are encodedwithout indicator As a result high embedding capacity andlow bit rate scenarios can be achieved As documented inthe experimental results the performance of the proposedscheme is significantly superior to that of the previous works
[10 12 15] in terms of the embedding rate bit rate andembedding capacity
The rest of this paper is organized as follows Section 2presents the related works to briefly introduce the conceptsof VQ and SMVQ Section 3 describes the proposed schemein detail Section 4 gives the experimental results and per-formances comparisons of the proposed scheme with formerapproaches Finally Section 5 draws conclusions
2 Backgrounds
21 Brief Concept of the VQ VQ initially involves codebookconstruction from a set of training images the trainingimages are partitioned into nonoverlapping blocks and themost representative blocks are selected to form codebook inwhich the elements in the codebook are called codewordsIn general the LBG algorithm [16] is employed to producethe desired codebook With the generated codebook eachblock in an image is encoded with the index of the nearestcodeword such that the total storage space for an imageis minimized To measure the similarity between a block119861 = (119887
1 1198872 119887
119896) and a codeword 119862
119895= (1198881198951 1198881198952 119888
119895119896)
where 119862119895is the 119895th codeword in the codebook the squared
Euclidean distance is generally used as below
119863(119861 119862119895) =
119896
sum
119894=1
(119887119894minus 119888119895119894)2
(1)
When the encoding process is completed block 119861 issimply represented by the index min 119895 where 119862min 119895 is thenearest codeword of the block 119861 For decoding process thetable lookup is performed with the same codebook as thatused by the encoder
22 SMVQ Kim initially proposed SMVQ for image coding[17] SMVQ exploits the high correlation among neighboringblocks and side-match prediction to create a state codebookwith a size smaller than that of the original VQ codebookThe side-match prediction assumes that the values of theadjacent pixels between the neighboring blocks are equal InSMVQ blocks located in the first row and first column ofan image are encoded by traditional VQ and the remainingblocks are predicted using the corresponding state codebookFigure 1 shows an example of the relationships among anencoding block119883 its upper neighboring block 119880 and its leftneighboring block 119871 for blocks of size 4times4This work denotesthe border vector and the side vector of block 119883 as 119861(119883) =
(11988311198832119883311988341198835119883911988313) = (1198871 1198872 1198873 1198874 1198875 1198876 1198877) and
119878(119883119880 119871) = ((11988013 + 1198714)2 11988014 11988015 11988016 1198718 11987112 11987116) =
(1199041 11990421199043 1199044 1199045 1199046 1199047) The SMD between a block 119883predicted
by blocks 119880 and 119871 and a codeword cw in the codebookis measured by the squared Euclidean distance denoted asfollows
SMD (119883 cw) = 119863 (S (XU L) B (cw)) =7
sum
119894=1
(119904119894minus 119887119894)2
(2)
For each block SMVQ sorts the codebook CB accordingto SMDs and thus the first 119865
119888codewords in the sorted
Mathematical Problems in Engineering 3
U13 U14 U15 U16L4 X1 X2 X3 X4L8 X5L12 X9L16 X13
Figure 1 An example of SMVQ
codebook are picked to form the state codebook SC in whichSMDs are the smallest for block 119883 Subsequently the SMVQsearches the closest codeword from the state codebook forblock 119883 and then the index of the closest codeword is usedto encode block 119883 Because the size of the state codebook issmaller than that of the VQ codebook the size of the SMVQindex is reduced to improve the coding gain However thereis a distortion between the block decoded by the SMVQindex and the corresponding block decoded by the VQindex
3 The Proposed Scheme
A grayscale cover image 119862 of size 119873 lowast 119872 is partitionedinto nonoverlapped blocks BLKi119894=1119871 of 119908 lowast ℎpixels inwhich 119871 = (119873119908) lowast (119872ℎ) Normally the variables 119908 =
ℎ = 4 Each block of the image can be encoded using VQand thus each block is represented with an index of thenearest codeword in the codebook AVQ-compressed imagedenoted by 119868 consists of indices The size of an index isproportional to the number of codewords in the codebookA SMVQ-compressed image is generated by applying SMVQpredictions to a VQ-compressed image To reconstruct a VQ-compressed image completely the size of the state codebookand that of the VQ codebook are first set as identical Nextthe codebook is sorted according to the SMDs which denotethe differences between an encoding block and the codewordsin the VQ codebook Finally the state codebook is generatedusing all of the codewords in the sorted codebook For a VQindex the corresponding SMVQ index can be obtained usingthe state codebook and no distortion is presented between thetwo codewords decoded by the two indices
Figures 2 and 3 show histograms of the VQ-compressedPeppers image and the corresponding SMVQ-compressedimage respectively with a codebook of size 256 Basedon the SMVQ predication most indices of the SMVQ-compressed image are distributed close to zero For a SMVQ-compressed image a very small value 119873
119911binary digits (119873
119911
bits) representing unsigned integers from 0 to 2119873119911
minus 1 cancover most of the indices In the proposed scheme the 2119873119911numbers are employed as indicators orand encoding codesTable 1 shows the functions of each number of the119873
119911binary
digits (119873119911bits) A binary number among 0 and 2
119873119911minus 3
represents an encoding code and indicator Binary numbers
0
100
200
300
400
500
600
700
800
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
Figure 2 Histogram of the VQ-compressed Peppers image
0
1000
2000
3000
4000
5000
6000
7000
8000
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
9000
Figure 3 Histogram of the SMVQ-compressed Peppers image
Table 1 The functions of each number of the 119873119911binary digits
(119873119911bits)
Number Function0sim2119873119911 minus 3 Indicator and encoding code2119873119911 minus 2 Indicator (indicator 0)
2119873119911 minus 1 Indicator (indicator 1)
2119873119911
minus2 and 2119873119911 minus1 represent indicators denoted by indicator0 and indicator 1 respectively
31 The Encoding Process Suppose the size of a secret datasd is 119904 bits During the encoding process a SMVQ indexwith a value less than 2
119873119911minus 2 which is defined as a tiny-
value index is denoted by119864119894 and the corresponding encoded
index is denoted by 1198641015840
119894 Then 1198641015840
119894is expressed in (3) where
and and denote exponential and concatenation operatorsrespectively
1198641015840
119894= 119864119894 sd (3)
4 Mathematical Problems in Engineering
Table 2 The ranges and symbols of tiny-value middle-value andlarge-value indices
Indices Range SymbolTiny-value indices 0sim2119873119911 minus 3 119864
119894
Middle-value indices 0 + 2119873119911minus 2sim2119873119888 minus 1 + 2
119873119911 minus 2 119866
119894
Large-value indices 2119873119888 + 2119873119911minus 2simcsminus 1 119874
119894
Most of the SMVQ indices can be encoded using (3)including no indicator and the rest are encoded with the helpof indicator 0 or indicator 1 However the adoption of theindicator increases the size of the encoded result A SMVQindex with a value in between (2
119873119911minus 2) + 0 and (2119873119911 minus 2) +
2119873119888minus1 is defined as amiddle-value index Indices withmiddle
values are encoded using indicator 0 concatenated with anencoding space with119873
119888bits Amiddle-value index is denoted
by119866119894 and the corresponding encoded index is denoted by1198661015840
119894
which can be expressed in the following
1198661015840
119894= 119894119899119889119894119888119886119905119900119903 0 119866
119894minus 2119873119911
+ 2 (4)
The119873119888-bit encoding space of (4) is not to hide data but to
collect more indices Indicator 1 is employed to assist encod-ing a SMVQ index which cannot be encoded using (3) or (4)A SMVQ index with a value more than (2119873119911 minus 2) + 2
119873119888minus 1
is defined as a large-value index denoted by 119874119894 and the
corresponding encoded index is denoted by 1198741015840119894which can be
expressed in the following
1198741015840
119894= 119894119899119889119894119888119886119905119900119903 1 119874
119894 (5)
Equation (5) not only hides no data but also increases anadditional indicator with 119873
119911bits For a SMVQ-compressed
image most indices with tiny values are encoded using (3)which hides 119904 bits and generates an encoded result with119873
119911+119904
bits for each index a few indices are encoded using (4)which generates an encoded result with119873
119911+119873119888bits for each
index few indices are encoded using (5) which generates anencoded result with119873
119911+log2(cs) bits for each index where cs
indicates the size of a codebook Table 2 shows the ranges andsymbols of tiny-value middle-value and large-value indicesThe size of an embedded image can be measured with (6)where 119909 denotes the size of 119909 and119868
119911 119868119888 and 119868
119900denote the
number of indices with tiny values middle values or largevalues respectively
1003817100381710038171003817embedded image1003817100381710038171003817 = (119873119911+ 119904) lowast 119868
119911
+ (119873119911+ 119873119888) lowast 119868119888
+ (119873119911+ log2(cs)) lowast 119868
119900
(6)
In (6) (119873119911+ 119904) and (119873
119911+ 119873119888) can be controlled in less
or more than log2(cs) bits and only the third item always
requires a space with more than log2(cs) bits It is remarkable
that a larger 119904 leads to a larger embedding capacity and viceversa Moreover a larger 119873
119888increases 119868
119888 while decreasing
119868119900 and vice versa Based on the special index distribution
property of the SMVQmost indices can be encodedusing (3)
An encoding index with value Q
Is Q in between(2Nz
minus 2) + 0 and (2Nzminus 2) + 2
Ncminus 1
A middle-valueindex encodedusing (4)
A large-valueindex encodedusing (5)
A tiny-valueindex encodedusing (3)
No
No
Yes
YesQ le 2
N119911 minus 3
Figure 4 An example of the index classification and encodingschemes
in which no indicator is required and 119904-bit data are hiddenin each index and thus a high embedding capacity and alow embedded image size can be achieved in the encodingprocess The algorithm of the embedding process is summa-rized as follows and an example of the index classification andencoding schemes is shown in Figure 4
Input A grayscale cover image 119862 of size119873 lowast119872 a codebookCB of size cs a secret data stream and parameters119873
119911and119873
119888
OutputThe code stream in binary form cs 119887 and parameters119873119911and119873
119888
Step 1 Compress the cover image 119862 using VQ to obtain theVQ-compressed image 119868 of size (119873119908) lowast (119872ℎ)
Step 2 Perform SMVQ predictions with the VQ-compressedimage 119868 to create the SMVQ-compressed image 119879 of size(119873119908) lowast (119872ℎ)
Step 3 Read the next SMVQ index denoted as 119876 from theSMVQ-compressed image 119879 with the raster scan order
Step 4 If 119876 le 2119873119911 minus 3 (a tiny-value index) then we have the
following
Step 41 Read 119904-bit data sd from the secret data stream
Step 42 119876 is encoded using (3) cs 119887 = cs 119887 D2B(119876) sdwhere denotes a concatenation operation and D2B denotesa decimal to binary operation
Step 5 If119876 is in between (2119873119911 minus 2) + 0 and (2119873119911 minus 2) + 2
119873119888 minus 1
(a middle-value index) then 119876 is encoded using (4) cs119887=
cs 119887 119894119899119889119894119888119886119905119900119903 0 D2B(119876 minus 2119873119911 + 2)
Mathematical Problems in Engineering 5
(a)
00000 01100 00110 01010 00010 01111 001100 12 6 10 2 15 6
(b) (c)
0 1 2 3
4 5 1 7
8 0 1 10
127 2 1 0 1 || 15
1 || 2
1 || 6
2 || 0
3 || 127
2 || 6
2 || 2
0 || 0
2 || 0
0 || 10
2 || 3
1 || 12
0 || 6
2 || 8
2 || 5
2 || 1
Figure 5 An example of the encoding process for hiding 5-bit secretdata (a) Secret data (b) SMVQ-compressed image and (c) encodedimage
Step 6
If 119876 ge (2119873119911 minus 2) + 2
119873119888 (a large-value index) then
119876 is encoded using (5)cs 119887 = cs 119887 119894119899119889119894119888119886119905119900119903 1 D2B(119876)
Step 7 Repeat Steps 3ndash7 until all of the indices in the SMVQ-compressed image are processed
Step 8 Output the code stream cs 119887 and parameters 119873119911and
119873119888
Figure 5 shows an example of the encoding process forhiding 5-bit secret data Let 119868
0be indicator 0 and 119868
1indicator
1 where 119873119911= 2 119873
119888= 5 119904 = 5 and 119888119904 = 128 and the size
of the state codebook is identical to that of the VQ codebookFor each image (Figure 5) let 119877
1to 1198774be the indices located
in the first row and 1198775to 1198778the indices located in the second
row and so on For the SMVQ-compressed image 1198771has a
value 0 and it is less than 2119873119911 minus2 The index is encoded using(3) Secret data are (00000)
2 According to (3) the index with
119873119911bits followed by secret data (00000)
2is presented as the
encoded result for 1198771 1198774has a value of 3 and it is in between
(2119873119911
minus 2) + 0 and (2119873119911
minus 2) + 2119873119888
minus 1 An indicator 1198680 2
followed by the offset index 1 is the encoded result for119877411987713
has a value of 127 and it is more than (2119873119911
minus 2) + 2119873119888
minus 1 Anindicator 119868
1 3 followed by the index 127 is the encoded result
for 11987713 All of the encoded results are in binary form and the
length of the encoded result of each index is (119873119911+119904) (119873
119911+119873119888)
or (119873119911+ log2(cs)) bits using (3) (4) or (5) respectively
32TheExtraction andReversion Process Thegoal of the dataextraction and reversion process is to extract the embeddedsecret data from an embedded image and then recover theoriginal VQ-compressed imageThe extraction and reversionprocess is introduced as below For an encoded index 119873
119911
bits are firstly read from code stream cs 119887 and the 119873119911-bit
binary value is converted to the decimal value denoted by119889 119873119911 In the first case if 119889 119873
119911is less than 2
119873119911minus 2 the
encoded index is an 119864119894index Then the original index is
directly restored by 119889 119873119911 Next read 119904 bits from the code
stream 119888 119887119904 to reconstruct the secret data In the second case
if 119889 119873119911is 2119873119911 minus 1 the encoded index is an 119874
119894index Then
read the next log2(cs) bits and convert them into decimal
value denoted by 119889 119874119894 the original index is recovered by
119889 119874119894 In the third case if 119889 119873
119911is 2119873119911 minus 2 the encoded index
is a 119866119894index Then read the next 119873
119888bits and convert them
into decimal value 119889 119866119894 the original index is recovered by
119889119866119894
+ (2119873119885 minus 2) When the original SMVQ-compressed image
has been reconstructed SMVQ predictions can be employedto restore the original VQ-compressed image In summarythe extraction and reversion process of the proposed schemeis organized as below
Input The code stream in binary form cs 119887 and parameters119873119911and119873
119888
Output The reconstructed VQ-compressed image 1198681015840 of size119873 lowast119872 and the retrieved secret data stream
Step 1 Set ptr = 1 and the restored sd = Null
Step 2 Read the next119873119911bits from the code stream 119888 119887119904 and
convert the119873119911bits to decimal value denoted by119889 119873
119911119889 119873119911=
B2D(cs 119887(ptr ptr + 119873119911minus 1) ptr = ptr + 119873
119911
Step 3 If 119889 119873119911lt 2119873119911
minus 2 (a tiny-value index) then
restored index = 119889 119873119911
read the next 119904 bits denoted by sd1015840 from code stream119888 119887119904restored sd = restored sd sd1015840ptr = ptr + 119904
Step 4 If 119889 119873119911= 2119873119911
minus 1 (indicator 1) then
read the next log2(cs) bits from the code stream c bs
119889 119874119894= B2D(cs 119887(ptr ptr + log
2(cs) minus 1) ptr = ptr +
log2(cs)
restored index = 119889 119874119894
Step 5 If 119889 119873119911= 2119873119911
minus 2 (indicator 0) then
read the next119873119888bits from the code stream 119888 119887119904
119889 119866119894= B2D(cs 119887(ptr ptr + 119873
119888minus 1) ptr = ptr + 119873
119888
restored index = 119889 119866119894+ (2119873119911
minus 2)
Step 6 Repeat Steps 2ndash6 until all of the bits in the code streamare processed
Step 7 Apply the SMVQ prediction to the reconstructedSMVQ-compressed image to obtain the original VQ-compressed image
Step 8 Output the reconstructed VQ-compressed image andthe restored secret data stream
4 Experimental Results
In this section we present the experimental results forevaluating the performance of the proposed scheme in terms
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 3
U13 U14 U15 U16L4 X1 X2 X3 X4L8 X5L12 X9L16 X13
Figure 1 An example of SMVQ
codebook are picked to form the state codebook SC in whichSMDs are the smallest for block 119883 Subsequently the SMVQsearches the closest codeword from the state codebook forblock 119883 and then the index of the closest codeword is usedto encode block 119883 Because the size of the state codebook issmaller than that of the VQ codebook the size of the SMVQindex is reduced to improve the coding gain However thereis a distortion between the block decoded by the SMVQindex and the corresponding block decoded by the VQindex
3 The Proposed Scheme
A grayscale cover image 119862 of size 119873 lowast 119872 is partitionedinto nonoverlapped blocks BLKi119894=1119871 of 119908 lowast ℎpixels inwhich 119871 = (119873119908) lowast (119872ℎ) Normally the variables 119908 =
ℎ = 4 Each block of the image can be encoded using VQand thus each block is represented with an index of thenearest codeword in the codebook AVQ-compressed imagedenoted by 119868 consists of indices The size of an index isproportional to the number of codewords in the codebookA SMVQ-compressed image is generated by applying SMVQpredictions to a VQ-compressed image To reconstruct a VQ-compressed image completely the size of the state codebookand that of the VQ codebook are first set as identical Nextthe codebook is sorted according to the SMDs which denotethe differences between an encoding block and the codewordsin the VQ codebook Finally the state codebook is generatedusing all of the codewords in the sorted codebook For a VQindex the corresponding SMVQ index can be obtained usingthe state codebook and no distortion is presented between thetwo codewords decoded by the two indices
Figures 2 and 3 show histograms of the VQ-compressedPeppers image and the corresponding SMVQ-compressedimage respectively with a codebook of size 256 Basedon the SMVQ predication most indices of the SMVQ-compressed image are distributed close to zero For a SMVQ-compressed image a very small value 119873
119911binary digits (119873
119911
bits) representing unsigned integers from 0 to 2119873119911
minus 1 cancover most of the indices In the proposed scheme the 2119873119911numbers are employed as indicators orand encoding codesTable 1 shows the functions of each number of the119873
119911binary
digits (119873119911bits) A binary number among 0 and 2
119873119911minus 3
represents an encoding code and indicator Binary numbers
0
100
200
300
400
500
600
700
800
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
Figure 2 Histogram of the VQ-compressed Peppers image
0
1000
2000
3000
4000
5000
6000
7000
8000
0 50 100 150 200 250 300Index value
Num
ber o
f ind
ices
9000
Figure 3 Histogram of the SMVQ-compressed Peppers image
Table 1 The functions of each number of the 119873119911binary digits
(119873119911bits)
Number Function0sim2119873119911 minus 3 Indicator and encoding code2119873119911 minus 2 Indicator (indicator 0)
2119873119911 minus 1 Indicator (indicator 1)
2119873119911
minus2 and 2119873119911 minus1 represent indicators denoted by indicator0 and indicator 1 respectively
31 The Encoding Process Suppose the size of a secret datasd is 119904 bits During the encoding process a SMVQ indexwith a value less than 2
119873119911minus 2 which is defined as a tiny-
value index is denoted by119864119894 and the corresponding encoded
index is denoted by 1198641015840
119894 Then 1198641015840
119894is expressed in (3) where
and and denote exponential and concatenation operatorsrespectively
1198641015840
119894= 119864119894 sd (3)
4 Mathematical Problems in Engineering
Table 2 The ranges and symbols of tiny-value middle-value andlarge-value indices
Indices Range SymbolTiny-value indices 0sim2119873119911 minus 3 119864
119894
Middle-value indices 0 + 2119873119911minus 2sim2119873119888 minus 1 + 2
119873119911 minus 2 119866
119894
Large-value indices 2119873119888 + 2119873119911minus 2simcsminus 1 119874
119894
Most of the SMVQ indices can be encoded using (3)including no indicator and the rest are encoded with the helpof indicator 0 or indicator 1 However the adoption of theindicator increases the size of the encoded result A SMVQindex with a value in between (2
119873119911minus 2) + 0 and (2119873119911 minus 2) +
2119873119888minus1 is defined as amiddle-value index Indices withmiddle
values are encoded using indicator 0 concatenated with anencoding space with119873
119888bits Amiddle-value index is denoted
by119866119894 and the corresponding encoded index is denoted by1198661015840
119894
which can be expressed in the following
1198661015840
119894= 119894119899119889119894119888119886119905119900119903 0 119866
119894minus 2119873119911
+ 2 (4)
The119873119888-bit encoding space of (4) is not to hide data but to
collect more indices Indicator 1 is employed to assist encod-ing a SMVQ index which cannot be encoded using (3) or (4)A SMVQ index with a value more than (2119873119911 minus 2) + 2
119873119888minus 1
is defined as a large-value index denoted by 119874119894 and the
corresponding encoded index is denoted by 1198741015840119894which can be
expressed in the following
1198741015840
119894= 119894119899119889119894119888119886119905119900119903 1 119874
119894 (5)
Equation (5) not only hides no data but also increases anadditional indicator with 119873
119911bits For a SMVQ-compressed
image most indices with tiny values are encoded using (3)which hides 119904 bits and generates an encoded result with119873
119911+119904
bits for each index a few indices are encoded using (4)which generates an encoded result with119873
119911+119873119888bits for each
index few indices are encoded using (5) which generates anencoded result with119873
119911+log2(cs) bits for each index where cs
indicates the size of a codebook Table 2 shows the ranges andsymbols of tiny-value middle-value and large-value indicesThe size of an embedded image can be measured with (6)where 119909 denotes the size of 119909 and119868
119911 119868119888 and 119868
119900denote the
number of indices with tiny values middle values or largevalues respectively
1003817100381710038171003817embedded image1003817100381710038171003817 = (119873119911+ 119904) lowast 119868
119911
+ (119873119911+ 119873119888) lowast 119868119888
+ (119873119911+ log2(cs)) lowast 119868
119900
(6)
In (6) (119873119911+ 119904) and (119873
119911+ 119873119888) can be controlled in less
or more than log2(cs) bits and only the third item always
requires a space with more than log2(cs) bits It is remarkable
that a larger 119904 leads to a larger embedding capacity and viceversa Moreover a larger 119873
119888increases 119868
119888 while decreasing
119868119900 and vice versa Based on the special index distribution
property of the SMVQmost indices can be encodedusing (3)
An encoding index with value Q
Is Q in between(2Nz
minus 2) + 0 and (2Nzminus 2) + 2
Ncminus 1
A middle-valueindex encodedusing (4)
A large-valueindex encodedusing (5)
A tiny-valueindex encodedusing (3)
No
No
Yes
YesQ le 2
N119911 minus 3
Figure 4 An example of the index classification and encodingschemes
in which no indicator is required and 119904-bit data are hiddenin each index and thus a high embedding capacity and alow embedded image size can be achieved in the encodingprocess The algorithm of the embedding process is summa-rized as follows and an example of the index classification andencoding schemes is shown in Figure 4
Input A grayscale cover image 119862 of size119873 lowast119872 a codebookCB of size cs a secret data stream and parameters119873
119911and119873
119888
OutputThe code stream in binary form cs 119887 and parameters119873119911and119873
119888
Step 1 Compress the cover image 119862 using VQ to obtain theVQ-compressed image 119868 of size (119873119908) lowast (119872ℎ)
Step 2 Perform SMVQ predictions with the VQ-compressedimage 119868 to create the SMVQ-compressed image 119879 of size(119873119908) lowast (119872ℎ)
Step 3 Read the next SMVQ index denoted as 119876 from theSMVQ-compressed image 119879 with the raster scan order
Step 4 If 119876 le 2119873119911 minus 3 (a tiny-value index) then we have the
following
Step 41 Read 119904-bit data sd from the secret data stream
Step 42 119876 is encoded using (3) cs 119887 = cs 119887 D2B(119876) sdwhere denotes a concatenation operation and D2B denotesa decimal to binary operation
Step 5 If119876 is in between (2119873119911 minus 2) + 0 and (2119873119911 minus 2) + 2
119873119888 minus 1
(a middle-value index) then 119876 is encoded using (4) cs119887=
cs 119887 119894119899119889119894119888119886119905119900119903 0 D2B(119876 minus 2119873119911 + 2)
Mathematical Problems in Engineering 5
(a)
00000 01100 00110 01010 00010 01111 001100 12 6 10 2 15 6
(b) (c)
0 1 2 3
4 5 1 7
8 0 1 10
127 2 1 0 1 || 15
1 || 2
1 || 6
2 || 0
3 || 127
2 || 6
2 || 2
0 || 0
2 || 0
0 || 10
2 || 3
1 || 12
0 || 6
2 || 8
2 || 5
2 || 1
Figure 5 An example of the encoding process for hiding 5-bit secretdata (a) Secret data (b) SMVQ-compressed image and (c) encodedimage
Step 6
If 119876 ge (2119873119911 minus 2) + 2
119873119888 (a large-value index) then
119876 is encoded using (5)cs 119887 = cs 119887 119894119899119889119894119888119886119905119900119903 1 D2B(119876)
Step 7 Repeat Steps 3ndash7 until all of the indices in the SMVQ-compressed image are processed
Step 8 Output the code stream cs 119887 and parameters 119873119911and
119873119888
Figure 5 shows an example of the encoding process forhiding 5-bit secret data Let 119868
0be indicator 0 and 119868
1indicator
1 where 119873119911= 2 119873
119888= 5 119904 = 5 and 119888119904 = 128 and the size
of the state codebook is identical to that of the VQ codebookFor each image (Figure 5) let 119877
1to 1198774be the indices located
in the first row and 1198775to 1198778the indices located in the second
row and so on For the SMVQ-compressed image 1198771has a
value 0 and it is less than 2119873119911 minus2 The index is encoded using(3) Secret data are (00000)
2 According to (3) the index with
119873119911bits followed by secret data (00000)
2is presented as the
encoded result for 1198771 1198774has a value of 3 and it is in between
(2119873119911
minus 2) + 0 and (2119873119911
minus 2) + 2119873119888
minus 1 An indicator 1198680 2
followed by the offset index 1 is the encoded result for119877411987713
has a value of 127 and it is more than (2119873119911
minus 2) + 2119873119888
minus 1 Anindicator 119868
1 3 followed by the index 127 is the encoded result
for 11987713 All of the encoded results are in binary form and the
length of the encoded result of each index is (119873119911+119904) (119873
119911+119873119888)
or (119873119911+ log2(cs)) bits using (3) (4) or (5) respectively
32TheExtraction andReversion Process Thegoal of the dataextraction and reversion process is to extract the embeddedsecret data from an embedded image and then recover theoriginal VQ-compressed imageThe extraction and reversionprocess is introduced as below For an encoded index 119873
119911
bits are firstly read from code stream cs 119887 and the 119873119911-bit
binary value is converted to the decimal value denoted by119889 119873119911 In the first case if 119889 119873
119911is less than 2
119873119911minus 2 the
encoded index is an 119864119894index Then the original index is
directly restored by 119889 119873119911 Next read 119904 bits from the code
stream 119888 119887119904 to reconstruct the secret data In the second case
if 119889 119873119911is 2119873119911 minus 1 the encoded index is an 119874
119894index Then
read the next log2(cs) bits and convert them into decimal
value denoted by 119889 119874119894 the original index is recovered by
119889 119874119894 In the third case if 119889 119873
119911is 2119873119911 minus 2 the encoded index
is a 119866119894index Then read the next 119873
119888bits and convert them
into decimal value 119889 119866119894 the original index is recovered by
119889119866119894
+ (2119873119885 minus 2) When the original SMVQ-compressed image
has been reconstructed SMVQ predictions can be employedto restore the original VQ-compressed image In summarythe extraction and reversion process of the proposed schemeis organized as below
Input The code stream in binary form cs 119887 and parameters119873119911and119873
119888
Output The reconstructed VQ-compressed image 1198681015840 of size119873 lowast119872 and the retrieved secret data stream
Step 1 Set ptr = 1 and the restored sd = Null
Step 2 Read the next119873119911bits from the code stream 119888 119887119904 and
convert the119873119911bits to decimal value denoted by119889 119873
119911119889 119873119911=
B2D(cs 119887(ptr ptr + 119873119911minus 1) ptr = ptr + 119873
119911
Step 3 If 119889 119873119911lt 2119873119911
minus 2 (a tiny-value index) then
restored index = 119889 119873119911
read the next 119904 bits denoted by sd1015840 from code stream119888 119887119904restored sd = restored sd sd1015840ptr = ptr + 119904
Step 4 If 119889 119873119911= 2119873119911
minus 1 (indicator 1) then
read the next log2(cs) bits from the code stream c bs
119889 119874119894= B2D(cs 119887(ptr ptr + log
2(cs) minus 1) ptr = ptr +
log2(cs)
restored index = 119889 119874119894
Step 5 If 119889 119873119911= 2119873119911
minus 2 (indicator 0) then
read the next119873119888bits from the code stream 119888 119887119904
119889 119866119894= B2D(cs 119887(ptr ptr + 119873
119888minus 1) ptr = ptr + 119873
119888
restored index = 119889 119866119894+ (2119873119911
minus 2)
Step 6 Repeat Steps 2ndash6 until all of the bits in the code streamare processed
Step 7 Apply the SMVQ prediction to the reconstructedSMVQ-compressed image to obtain the original VQ-compressed image
Step 8 Output the reconstructed VQ-compressed image andthe restored secret data stream
4 Experimental Results
In this section we present the experimental results forevaluating the performance of the proposed scheme in terms
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Mathematical Problems in Engineering
Table 2 The ranges and symbols of tiny-value middle-value andlarge-value indices
Indices Range SymbolTiny-value indices 0sim2119873119911 minus 3 119864
119894
Middle-value indices 0 + 2119873119911minus 2sim2119873119888 minus 1 + 2
119873119911 minus 2 119866
119894
Large-value indices 2119873119888 + 2119873119911minus 2simcsminus 1 119874
119894
Most of the SMVQ indices can be encoded using (3)including no indicator and the rest are encoded with the helpof indicator 0 or indicator 1 However the adoption of theindicator increases the size of the encoded result A SMVQindex with a value in between (2
119873119911minus 2) + 0 and (2119873119911 minus 2) +
2119873119888minus1 is defined as amiddle-value index Indices withmiddle
values are encoded using indicator 0 concatenated with anencoding space with119873
119888bits Amiddle-value index is denoted
by119866119894 and the corresponding encoded index is denoted by1198661015840
119894
which can be expressed in the following
1198661015840
119894= 119894119899119889119894119888119886119905119900119903 0 119866
119894minus 2119873119911
+ 2 (4)
The119873119888-bit encoding space of (4) is not to hide data but to
collect more indices Indicator 1 is employed to assist encod-ing a SMVQ index which cannot be encoded using (3) or (4)A SMVQ index with a value more than (2119873119911 minus 2) + 2
119873119888minus 1
is defined as a large-value index denoted by 119874119894 and the
corresponding encoded index is denoted by 1198741015840119894which can be
expressed in the following
1198741015840
119894= 119894119899119889119894119888119886119905119900119903 1 119874
119894 (5)
Equation (5) not only hides no data but also increases anadditional indicator with 119873
119911bits For a SMVQ-compressed
image most indices with tiny values are encoded using (3)which hides 119904 bits and generates an encoded result with119873
119911+119904
bits for each index a few indices are encoded using (4)which generates an encoded result with119873
119911+119873119888bits for each
index few indices are encoded using (5) which generates anencoded result with119873
119911+log2(cs) bits for each index where cs
indicates the size of a codebook Table 2 shows the ranges andsymbols of tiny-value middle-value and large-value indicesThe size of an embedded image can be measured with (6)where 119909 denotes the size of 119909 and119868
119911 119868119888 and 119868
119900denote the
number of indices with tiny values middle values or largevalues respectively
1003817100381710038171003817embedded image1003817100381710038171003817 = (119873119911+ 119904) lowast 119868
119911
+ (119873119911+ 119873119888) lowast 119868119888
+ (119873119911+ log2(cs)) lowast 119868
119900
(6)
In (6) (119873119911+ 119904) and (119873
119911+ 119873119888) can be controlled in less
or more than log2(cs) bits and only the third item always
requires a space with more than log2(cs) bits It is remarkable
that a larger 119904 leads to a larger embedding capacity and viceversa Moreover a larger 119873
119888increases 119868
119888 while decreasing
119868119900 and vice versa Based on the special index distribution
property of the SMVQmost indices can be encodedusing (3)
An encoding index with value Q
Is Q in between(2Nz
minus 2) + 0 and (2Nzminus 2) + 2
Ncminus 1
A middle-valueindex encodedusing (4)
A large-valueindex encodedusing (5)
A tiny-valueindex encodedusing (3)
No
No
Yes
YesQ le 2
N119911 minus 3
Figure 4 An example of the index classification and encodingschemes
in which no indicator is required and 119904-bit data are hiddenin each index and thus a high embedding capacity and alow embedded image size can be achieved in the encodingprocess The algorithm of the embedding process is summa-rized as follows and an example of the index classification andencoding schemes is shown in Figure 4
Input A grayscale cover image 119862 of size119873 lowast119872 a codebookCB of size cs a secret data stream and parameters119873
119911and119873
119888
OutputThe code stream in binary form cs 119887 and parameters119873119911and119873
119888
Step 1 Compress the cover image 119862 using VQ to obtain theVQ-compressed image 119868 of size (119873119908) lowast (119872ℎ)
Step 2 Perform SMVQ predictions with the VQ-compressedimage 119868 to create the SMVQ-compressed image 119879 of size(119873119908) lowast (119872ℎ)
Step 3 Read the next SMVQ index denoted as 119876 from theSMVQ-compressed image 119879 with the raster scan order
Step 4 If 119876 le 2119873119911 minus 3 (a tiny-value index) then we have the
following
Step 41 Read 119904-bit data sd from the secret data stream
Step 42 119876 is encoded using (3) cs 119887 = cs 119887 D2B(119876) sdwhere denotes a concatenation operation and D2B denotesa decimal to binary operation
Step 5 If119876 is in between (2119873119911 minus 2) + 0 and (2119873119911 minus 2) + 2
119873119888 minus 1
(a middle-value index) then 119876 is encoded using (4) cs119887=
cs 119887 119894119899119889119894119888119886119905119900119903 0 D2B(119876 minus 2119873119911 + 2)
Mathematical Problems in Engineering 5
(a)
00000 01100 00110 01010 00010 01111 001100 12 6 10 2 15 6
(b) (c)
0 1 2 3
4 5 1 7
8 0 1 10
127 2 1 0 1 || 15
1 || 2
1 || 6
2 || 0
3 || 127
2 || 6
2 || 2
0 || 0
2 || 0
0 || 10
2 || 3
1 || 12
0 || 6
2 || 8
2 || 5
2 || 1
Figure 5 An example of the encoding process for hiding 5-bit secretdata (a) Secret data (b) SMVQ-compressed image and (c) encodedimage
Step 6
If 119876 ge (2119873119911 minus 2) + 2
119873119888 (a large-value index) then
119876 is encoded using (5)cs 119887 = cs 119887 119894119899119889119894119888119886119905119900119903 1 D2B(119876)
Step 7 Repeat Steps 3ndash7 until all of the indices in the SMVQ-compressed image are processed
Step 8 Output the code stream cs 119887 and parameters 119873119911and
119873119888
Figure 5 shows an example of the encoding process forhiding 5-bit secret data Let 119868
0be indicator 0 and 119868
1indicator
1 where 119873119911= 2 119873
119888= 5 119904 = 5 and 119888119904 = 128 and the size
of the state codebook is identical to that of the VQ codebookFor each image (Figure 5) let 119877
1to 1198774be the indices located
in the first row and 1198775to 1198778the indices located in the second
row and so on For the SMVQ-compressed image 1198771has a
value 0 and it is less than 2119873119911 minus2 The index is encoded using(3) Secret data are (00000)
2 According to (3) the index with
119873119911bits followed by secret data (00000)
2is presented as the
encoded result for 1198771 1198774has a value of 3 and it is in between
(2119873119911
minus 2) + 0 and (2119873119911
minus 2) + 2119873119888
minus 1 An indicator 1198680 2
followed by the offset index 1 is the encoded result for119877411987713
has a value of 127 and it is more than (2119873119911
minus 2) + 2119873119888
minus 1 Anindicator 119868
1 3 followed by the index 127 is the encoded result
for 11987713 All of the encoded results are in binary form and the
length of the encoded result of each index is (119873119911+119904) (119873
119911+119873119888)
or (119873119911+ log2(cs)) bits using (3) (4) or (5) respectively
32TheExtraction andReversion Process Thegoal of the dataextraction and reversion process is to extract the embeddedsecret data from an embedded image and then recover theoriginal VQ-compressed imageThe extraction and reversionprocess is introduced as below For an encoded index 119873
119911
bits are firstly read from code stream cs 119887 and the 119873119911-bit
binary value is converted to the decimal value denoted by119889 119873119911 In the first case if 119889 119873
119911is less than 2
119873119911minus 2 the
encoded index is an 119864119894index Then the original index is
directly restored by 119889 119873119911 Next read 119904 bits from the code
stream 119888 119887119904 to reconstruct the secret data In the second case
if 119889 119873119911is 2119873119911 minus 1 the encoded index is an 119874
119894index Then
read the next log2(cs) bits and convert them into decimal
value denoted by 119889 119874119894 the original index is recovered by
119889 119874119894 In the third case if 119889 119873
119911is 2119873119911 minus 2 the encoded index
is a 119866119894index Then read the next 119873
119888bits and convert them
into decimal value 119889 119866119894 the original index is recovered by
119889119866119894
+ (2119873119885 minus 2) When the original SMVQ-compressed image
has been reconstructed SMVQ predictions can be employedto restore the original VQ-compressed image In summarythe extraction and reversion process of the proposed schemeis organized as below
Input The code stream in binary form cs 119887 and parameters119873119911and119873
119888
Output The reconstructed VQ-compressed image 1198681015840 of size119873 lowast119872 and the retrieved secret data stream
Step 1 Set ptr = 1 and the restored sd = Null
Step 2 Read the next119873119911bits from the code stream 119888 119887119904 and
convert the119873119911bits to decimal value denoted by119889 119873
119911119889 119873119911=
B2D(cs 119887(ptr ptr + 119873119911minus 1) ptr = ptr + 119873
119911
Step 3 If 119889 119873119911lt 2119873119911
minus 2 (a tiny-value index) then
restored index = 119889 119873119911
read the next 119904 bits denoted by sd1015840 from code stream119888 119887119904restored sd = restored sd sd1015840ptr = ptr + 119904
Step 4 If 119889 119873119911= 2119873119911
minus 1 (indicator 1) then
read the next log2(cs) bits from the code stream c bs
119889 119874119894= B2D(cs 119887(ptr ptr + log
2(cs) minus 1) ptr = ptr +
log2(cs)
restored index = 119889 119874119894
Step 5 If 119889 119873119911= 2119873119911
minus 2 (indicator 0) then
read the next119873119888bits from the code stream 119888 119887119904
119889 119866119894= B2D(cs 119887(ptr ptr + 119873
119888minus 1) ptr = ptr + 119873
119888
restored index = 119889 119866119894+ (2119873119911
minus 2)
Step 6 Repeat Steps 2ndash6 until all of the bits in the code streamare processed
Step 7 Apply the SMVQ prediction to the reconstructedSMVQ-compressed image to obtain the original VQ-compressed image
Step 8 Output the reconstructed VQ-compressed image andthe restored secret data stream
4 Experimental Results
In this section we present the experimental results forevaluating the performance of the proposed scheme in terms
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 5
(a)
00000 01100 00110 01010 00010 01111 001100 12 6 10 2 15 6
(b) (c)
0 1 2 3
4 5 1 7
8 0 1 10
127 2 1 0 1 || 15
1 || 2
1 || 6
2 || 0
3 || 127
2 || 6
2 || 2
0 || 0
2 || 0
0 || 10
2 || 3
1 || 12
0 || 6
2 || 8
2 || 5
2 || 1
Figure 5 An example of the encoding process for hiding 5-bit secretdata (a) Secret data (b) SMVQ-compressed image and (c) encodedimage
Step 6
If 119876 ge (2119873119911 minus 2) + 2
119873119888 (a large-value index) then
119876 is encoded using (5)cs 119887 = cs 119887 119894119899119889119894119888119886119905119900119903 1 D2B(119876)
Step 7 Repeat Steps 3ndash7 until all of the indices in the SMVQ-compressed image are processed
Step 8 Output the code stream cs 119887 and parameters 119873119911and
119873119888
Figure 5 shows an example of the encoding process forhiding 5-bit secret data Let 119868
0be indicator 0 and 119868
1indicator
1 where 119873119911= 2 119873
119888= 5 119904 = 5 and 119888119904 = 128 and the size
of the state codebook is identical to that of the VQ codebookFor each image (Figure 5) let 119877
1to 1198774be the indices located
in the first row and 1198775to 1198778the indices located in the second
row and so on For the SMVQ-compressed image 1198771has a
value 0 and it is less than 2119873119911 minus2 The index is encoded using(3) Secret data are (00000)
2 According to (3) the index with
119873119911bits followed by secret data (00000)
2is presented as the
encoded result for 1198771 1198774has a value of 3 and it is in between
(2119873119911
minus 2) + 0 and (2119873119911
minus 2) + 2119873119888
minus 1 An indicator 1198680 2
followed by the offset index 1 is the encoded result for119877411987713
has a value of 127 and it is more than (2119873119911
minus 2) + 2119873119888
minus 1 Anindicator 119868
1 3 followed by the index 127 is the encoded result
for 11987713 All of the encoded results are in binary form and the
length of the encoded result of each index is (119873119911+119904) (119873
119911+119873119888)
or (119873119911+ log2(cs)) bits using (3) (4) or (5) respectively
32TheExtraction andReversion Process Thegoal of the dataextraction and reversion process is to extract the embeddedsecret data from an embedded image and then recover theoriginal VQ-compressed imageThe extraction and reversionprocess is introduced as below For an encoded index 119873
119911
bits are firstly read from code stream cs 119887 and the 119873119911-bit
binary value is converted to the decimal value denoted by119889 119873119911 In the first case if 119889 119873
119911is less than 2
119873119911minus 2 the
encoded index is an 119864119894index Then the original index is
directly restored by 119889 119873119911 Next read 119904 bits from the code
stream 119888 119887119904 to reconstruct the secret data In the second case
if 119889 119873119911is 2119873119911 minus 1 the encoded index is an 119874
119894index Then
read the next log2(cs) bits and convert them into decimal
value denoted by 119889 119874119894 the original index is recovered by
119889 119874119894 In the third case if 119889 119873
119911is 2119873119911 minus 2 the encoded index
is a 119866119894index Then read the next 119873
119888bits and convert them
into decimal value 119889 119866119894 the original index is recovered by
119889119866119894
+ (2119873119885 minus 2) When the original SMVQ-compressed image
has been reconstructed SMVQ predictions can be employedto restore the original VQ-compressed image In summarythe extraction and reversion process of the proposed schemeis organized as below
Input The code stream in binary form cs 119887 and parameters119873119911and119873
119888
Output The reconstructed VQ-compressed image 1198681015840 of size119873 lowast119872 and the retrieved secret data stream
Step 1 Set ptr = 1 and the restored sd = Null
Step 2 Read the next119873119911bits from the code stream 119888 119887119904 and
convert the119873119911bits to decimal value denoted by119889 119873
119911119889 119873119911=
B2D(cs 119887(ptr ptr + 119873119911minus 1) ptr = ptr + 119873
119911
Step 3 If 119889 119873119911lt 2119873119911
minus 2 (a tiny-value index) then
restored index = 119889 119873119911
read the next 119904 bits denoted by sd1015840 from code stream119888 119887119904restored sd = restored sd sd1015840ptr = ptr + 119904
Step 4 If 119889 119873119911= 2119873119911
minus 1 (indicator 1) then
read the next log2(cs) bits from the code stream c bs
119889 119874119894= B2D(cs 119887(ptr ptr + log
2(cs) minus 1) ptr = ptr +
log2(cs)
restored index = 119889 119874119894
Step 5 If 119889 119873119911= 2119873119911
minus 2 (indicator 0) then
read the next119873119888bits from the code stream 119888 119887119904
119889 119866119894= B2D(cs 119887(ptr ptr + 119873
119888minus 1) ptr = ptr + 119873
119888
restored index = 119889 119866119894+ (2119873119911
minus 2)
Step 6 Repeat Steps 2ndash6 until all of the bits in the code streamare processed
Step 7 Apply the SMVQ prediction to the reconstructedSMVQ-compressed image to obtain the original VQ-compressed image
Step 8 Output the reconstructed VQ-compressed image andthe restored secret data stream
4 Experimental Results
In this section we present the experimental results forevaluating the performance of the proposed scheme in terms
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Mathematical Problems in Engineering
(a) Airplane (b) Lena
(c) Peppers (d) Boat
(e) Sailboat (f) Tiffany
(g) House (h) Elaine
(i) Zelda (j) Gold hill
Figure 6 Test images
of the embedding rate bit rate and embedding capacityVarious gray-level images as shown in Figure 6 are employedas the training images or cover images each of which is ofsize 512 lowast 512 The codebooks of sizes 128 256 512 and1024 with codewords of 16 dimensions were trained by theLBG algorithmwith ldquoAirplanerdquo ldquoBoatrdquo ldquoLenardquo ldquoPeppersrdquo andldquoSailboatrdquo as the training images
The secret data in the experiment is in binary format 0and 1 and which are generated by a pseudorandom numbergenerator If a high-level security is required secret data canbe encrypted prior to the embedding using the well-knowncryptographicmethods such as DES or RSA In this work theembedding rate (ER) bit rate (BR) and embedding capacity(EC) as defined in (7)ndash(9) respectively are employed toevaluate the performance of various schemes In general adata-hidingmethod with a small value in BR and large valuesin EC and ER endorses a good performance and vice versa
BR =
1003817100381710038171003817embedded image1003817100381710038171003817119873 lowast119872
(7)
EC = 119904 lowast 119868119911 (8)
ER =EC
1003817100381710038171003817embedded image1003817100381710038171003817 (9)
41 Experiments on Selecting the Appropriate Parameter N119911
To obtain the best performance of the proposed schemethe parameter 119873
119911is needed to be properly determined As
mentioned when 119873119911and 119904 are set to a larger value 119868
119911
EC and BR are increased The encoding length of a VQindex is log
2(cs) bits A VQ index encoded as an 119864
119894or 119866119894
index yields an additional space with log2(cs) minus 119873
119911bits
the space can be used to hide secret data or collect moreindices Experiments are performed to choose an appropriate119873119911value which can achieve the highest embedding rate for
different codebooks and different images First of all 119904 and119873119888are set to log
2(cs) minus 119873
119911 Figure 7 shows the performance
of the proposed algorithmusing different119873119911 codebook sizes
and test images For codebooks of sizes 128 and 256 SMVQprovides accurate prediction and the most of the VQ indicesare predicted by the codewords with smaller indices in theSMVQ state codebook In these cases the highest ERs can beobtained when119873
119911is set to a small value 3 For codebooks of
sizes 512 and 1024 119873119911must be set at greater values 4 or 5
to obtain a high ER For each image larger ERs appear as119873119911
with values of 3 3 4 and 5 for codebooks of sizes 128 256512 and 1024 respectively
The experiments are performed based on the selectedvarious119873
119911with the four codebooks of sizes 128 256 512 and
1024 Experimental results for the test images are shown inTable 3 in which different columns of the simulation resultsare obtained with (cs 119873
119911 119873119888 119904) = (128 3 4 4) (256 3
5 5) (512 4 5 5) and (1024 5 5 5) respectively FromTable 3 it can be seen that the BRs of the proposed schemeare only somewhat larger than that of the original VQ BRsand the embedding capacities of the proposed scheme arevery high Moreover for most images the high embeddingcapacities are still obtained when the sizes of codebooksare increased The smooth image such as ldquoTiffanyrdquo has ahigh embedding capacity than that of the complicated imagesuch as ldquoBoatrdquo since the SMVQ prediction normally presentsgood performance for a smooth image thus many indices areencoded with 119864
119894indices With the proposed method when
the codebooks are of sizes 128 256 512 and 1024 the averageECs are 3246 3637 3848 and 3872 bitsindex respectively
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 7
01
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(a)
02
03
04
05
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(b)
015
025
035
045
ER
2 3 4 5 6Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(c)
01
02
03
04
ER
2 3 4 5 6
Nz
Airplane
LenaPepper
BoatSailboat
TiffanyHouse
Elaine
(d)
Figure 7 The performance of the proposed algorithm using different 119873119911 different test images and different codebook sizes (a) codebook
size 128 (b) codebook size 256 (c) codebook size 512 and (d) codebook size 1024
In addition for applications of secret communication 119904 canbe set to a value smaller (larger) than log
2(cs) to obtain a
lower (higher) embedding capacity and consequently thecorresponding bit rate can be changed as well
42 Experiments on Comparing the Proposed Method withOther Methods To demonstrate the superiority of the pro-posed algorithm it is compared with the VQ-based schemeproposed by Chang et al [10] the VQSMVQ-based scheme
proposed by Yang and Lin [15] and SOC-based schemeproposed by Shie and Lin [12] as detailed below To ensureall of them can be fairly compared the proposed scheme isperformed with the 119873
119911 119873119888 and 119904 values are set to obtain an
embedding capacity close to that of the other schemes andthe codebook size for each comparison is set as identical
The results on the left and the middle of Table 4 showthe ERs BRs and ECs of the schemes developed by Changet al [10] and Yang and Lin [15] respectively The results
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Mathematical Problems in Engineering
Table 3 Simulation results for various test images and codebook sizes
Images
Codebook sizesVQ BR (in bpp)1280438 256050 5120563 10240625
119873119911119904
34 35 45 55ER
BR (in bpp)EC (in bits)
Airplane0454 0466 0438 03980446 0509 0574 064053144 62095 65905 66785
Lena0478 0471 0441 04030442 0505 0570 063755396 62350 65955 67390
Pepper0483 0483 0456 04120442 0504 0570 063655928 63810 68030 68685
Boat0423 0380 0354 03180446 0509 0579 065049432 50775 53720 54155
Sailboat0435 0416 0391 03440446 0509 0576 064550792 55515 59120 58110
Tiffany0492 0521 0473 04220441 0503 0569 063556860 68625 70475 70220
House0394 0370 0345 03140449 0513 0584 065446392 49780 52745 53840
Elaine0498 0484 0460 04120440 0503 0568 063457492 63775 68445 68385
on the right of Table 4 show the ERs BRs and ECs of theproposed scheme In Table 4 Chang et alrsquos scheme is VQ-based Yang and Linrsquos scheme is VQSMVQ-based and theproposed scheme is SMVQ-based First of all119873
119911119873119888 and 119904 of
the proposed scheme are set to obtain an embedding capacityclose to that of the other schemesThus when the codebooksare of sizes 128 256 512 and 1024 (119873
119911 119904 119873119888) are set at (2
3 log2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911) (33 log
2(cs) minus 119873
119911)
and (43 log2(cs) minus 119873
119911) respectively SMVQ can relocate
the histogram of the VQ indices to locations near zero Inthe proposed scheme the indices with values close to zeroare encoded using short codes without any indicator whilethe indices of values away from zero are encoded using theoriginal indexwith an indicator Based on the characteristic ofthe SMVQmost of the indices are with small values and onlya few indices have large values Thus this property leads tolow bit rate and high embedding capacity with the proposed
scheme In Chang et alrsquos [10] scheme when the numberof embeddable indices is increased the embedding capacityof each index is decreased and vice versa The satisfactoryEC cannot yield Moreover most indices are encoded by anindicator concatenated with the encoding code The lengthof an indicator or an encoding code is identical to that ofthe original VQ index The bit rate of an embedded imageis thus increased In Yang and Linrsquos [15] scheme a 1-bitindicator is always used to distinguish a SMVQ encodingfrom a VQ encoding and other indicators are required forVQ encoding leading to a high bit rate result Comparedwiththe other schemes the proposed scheme has low bit rate yethigh embedding capacity and embedding rate In particularwhen the embedding capacities of the proposed scheme arecontrolled to values close to that of the other schemes theproposed scheme presents much lower BRs even smallerthan that of the original VQ BRs for all of the test images
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 9
Table 4 Performance comparisons among Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme and the proposed scheme with various testimages and codebook sizes
Images
Chang et alrsquos [10] scheme Yang and Linrsquos [15] scheme The proposed schemeCodebook sizes Codebook sizesVQ BR
128 256 512 1024 256 5121280438 2560500 5120563 10240625
119873119911119904
23 33 33 43ER
BR (in bpp)EC (in bits)
Airplane0182 0149 0124 0119 0236 0208 0326 0343 0291 02690580 0670 0760 0850 0500 0550 0364 0414 0442 050327684 26170 24608 26432 30894 29954 31077 37257 33747 35448
Lena0179 0136 0115 0094 0234 0208 0332 0348 0275 02610580 0670 0750 0830 0500 0550 0360 0410 0445 050427164 23884 22626 20502 30736 29958 31308 37410 32067 34428
Peppers0187 0150 0116 0104 0244 0216 0338 0359 0287 02640580 0670 0740 0830 0490 0540 0359 0407 0440 050128372 26284 22592 22676 31346 30576 31743 38286 33180 34677
Toys0195 0160 0136 0119 0254 0234 0397 0373 0332 03020590 0680 0770 0860 0480 0510 0348 0405 0426 048730202 28606 27460 26760 31912 31226 36300 39594 37107 38529
Gold hill0142 0109 0084 0074 0203 0158 0306 0321 0248 02320590 0680 0750 0830 0540 0640 0364 0415 0454 051421928 19420 16548 16116 28792 26502 29211 34866 29451 31287
Zelda0179 0135 0111 0088 0250 0220 0338 0375 0308 02980580 0670 0740 0820 0480 0530 0357 0401 0431 048527258 23754 21506 18842 31444 30634 31704 39462 34809 37842
The results on the left of Table 5 show the ERs BRsand ECs of the SOC-based scheme with codebooks ofsizes 256 and 512 respectively developed by Shie and Lin[12] The results on the right of the Table 5 show the ERsBRs and ECs of the proposed schemewith codebooks of sizes256 and 512 respectively In Shie and Linrsquos [12] scheme a 1-bit indicator is always used to distinguish a SOC index froman OIV index in which a SOC is encoded using 1 + 119889 bitswhile an OIV index is encoded using 1 + log
2(cs) bits The
variable 119889 indicates the length of a search path and which isset to 2 to yield a better performance The proposed schemeencodes 119864
119894and 119866
119894indices using 119873
119911and log
2(cs) minus 119873
119911bits
where 119873119911is set to 3 to obtain embedding capacities close to
that of Shie and Linrsquos method The length of a VQ index islog2(cs) in which 119873
119911or 1 + 119889 bits are used to encode an 119864
119894
index or a SOC index and the remaining bits can be used tohide secret data In Table 5 first of all the encoded result foreach119864
119894or SOC index of the two schemes is set to log
2(cs) bits
which yields a hiding space of log2(cs) minus 1 minus 2 bits for a SOC
index and log2(cs)minus119873
119911bits for a119864
119894index For each test image
with a specific codebook size the bold-number columns inTable 5 show that the EC values of the proposed scheme are
significantly greater than that of Shie and Linrsquos scheme whilelower BR values and higher ER values can be achieved withthe proposed scheme
Next the hiding space of Shie and Linrsquos [12] scheme is setto log
2(cs) bits which can achieve a high embedding space
and the length of the encoded result for a SOC index is 1+ 2 + log
2(cs) bits Because the proposed scheme has good
performance in terms of ER BR and EC 119873119911 119873119888 and the
hiding space of the proposed scheme is set to 3 log2(cs) minus119873
119911
and log2(cs) minus 1 bits the set can offer an EC close to that of
Shie and Linrsquos scheme and decrease the length of the encodedresult of an119864
119894index to 3+ log
2(cs)minus1 bits For each codebook
size and test image the unmarked columns in Table 5 showthat the EC values of the proposed scheme are still muchgreater than that of Shie and Linrsquos scheme under the highhiding space while the proposed scheme can preserve lowBRvalues and high ER values Notably the ldquoZeldardquo and ldquoGoldhillrdquo images are not the training members for proper 119873
119911
selection Yet it till can achieve superior performancesAs mentioned the proposed scheme outperforms the
VQ-based scheme proposed by Chang et al [10] theVQSMVQ-based scheme proposed by Yang and Lin [15]
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Mathematical Problems in Engineering
Table 5 Performance comparisons between Shie and Linrsquos [12] scheme and the proposed scheme with various test images and codebooksizes
Images
Shie and Linrsquos [12] scheme The proposed schemeCodebook sizesVQ BR
2560500 5120563 2560500 5120563 2560500 5120563 2560500 5120563d119904encoded bits 119873
119911119904encoded bits
258 269 2811 2912 358 369 3710 3811ER
BR (in bpp)EC (in bits)
Airplane0383 0342 0499 0438 0466 0451 0550 05220522 0591 0643 0692 0509 0571 0603 065752500 52998 84000 79497 62095 67494 86933 89992
Lena0371 0308 0485 0400 0471 0431 0555 05030524 0594 0640 0686 0505 0567 0600 064950915 48000 81464 72000 62350 64134 87290 85512
Peppers0369 0298 0484 0389 0483 0447 0567 05180524 0595 0640 0684 0504 0567 0601 065150735 46506 81176 69759 63810 66360 89334 88480
Boat0272 0194 0374 0265 0380 0305 0462 03690533 0605 0621 0664 0509 0573 0587 063138015 30744 60824 46116 50775 45756 71085 61008
Sailboat0296 0214 0402 0289 0416 0347 0499 04140531 0604 0625 0668 0509 0571 0594 063841235 33792 65976 50688 55515 51942 77721 69256
Gold hill0338 0290 0450 0380 0440 0397 0524 04670527 0596 0634 0683 0503 0566 0592 064146695 45378 74712 68067 58110 58902 81354 78536
Zelda0361 0304 0475 0395 0500 0471 0584 05430525 0595 0638 0685 0502 0564 0602 065349665 47340 79464 71010 65770 69618 92078 92824
and SOC-based scheme proposed by Shie and Lin [12] inthree aspects that is embedding rate bit rate and embeddingcapacity Thus the proposed method can be considered aseffective coding candidate for reversible data hiding applica-tion
5 Conclusions
In this paper a novel reversible data-hiding scheme forSMVQ-compressed images is proposed In most VQSMVQhiding schemes the bit rate is increased when an indicatoris required to encode an index The proposed scheme appliesthe SMVQ prediction to relocate most of the indices of a VQ-compressed image to locations close to zero Consequentlythese small-value frequently used indices are encoded usingfewer bits and no indicator is required which thus increasesthe embedding capacity and decreases the bit rate Theexperimental results demonstrate that the proposed schemecan completely extract the secret data and recover the orig-inal VQ-compressed image Moreover comparing to otherschemes the proposed scheme can achieve better efficiencyunder each codebook size for various test images
Acknowledgment
Theworkwas supported byNational ScienceCouncil Repub-lic of China Taiwan under Grant NSC100-2221-E-182-047-MY3
References
[1] J Tian ldquoReversible data embedding using a difference expan-sionrdquo IEEE Transactions on Circuits and Systems for VideoTechnology vol 13 no 8 pp 890ndash896 2003
[2] Z Ni Y-Q Shi N Ansari and W Su ldquoReversible data hidingrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 3 pp 354ndash361 2006
[3] Y Hu H-K Lee and J Li ldquoDE-based reversible data hidingwith improved overflow location maprdquo IEEE Transactions onCircuits and Systems for Video Technology vol 19 no 2 pp 250ndash260 2009
[4] CH Lin andC Y Yang ldquoMultipurposewatermarking based onblind vector quantization (BVQ)rdquo Journal of InformationHidingand Multimedia Signal Processing vol 2 no 2 pp 239ndash2462011
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 11
[5] S Weng S C Chu N Cai and R Zhan ldquoInvariability of meanvalue based reversible watermarkingrdquo Journal of InformationHiding andMultimedia Signal Processing vol 4 no 2 pp 90ndash982013
[6] Y Wang J F Doherty and R E Van Dyck ldquoA wavelet-basedwatermarking algorithm for ownership verification of digitalimagesrdquo IEEE Transactions on Image Processing vol 11 no 2pp 77ndash88 2002
[7] J-D Lee Y-H Chiou and J-M Guo ldquoReversible data hidingbased on histogram modification of SMVQ indicesrdquo IEEETransactions on Information Forensics and Security vol 5 no4 pp 638ndash648 2010
[8] C Y Yang C H Lin and W C Hu ldquoReversible data hidingfor high-quality images based on integer wavelet transformrdquoJournal of InformationHiding andMultimedia Signal Processingvol 3 no 2 pp 142ndash150 2012
[9] M Jo andH Kim ldquoA digital image watermarking scheme basedon vector quantisationrdquo IEICE Transactions on Information andSystems vol E85-D no 6 pp 1054ndash1056 2002
[10] C-C Chang W-C Wu and Y-C Hu ldquoLossless recovery of aVQ index table with embedded secret datardquo Journal of VisualCommunication and Image Representation vol 18 no 3 pp207ndash216 2007
[11] C-C Chang G-M Chen and M-H Lin ldquoInformation hidingbased on search-order coding for VQ indicesrdquo Pattern Recogni-tion Letters vol 25 no 11 pp 1253ndash1261 2004
[12] S-C Shie and S D Lin ldquoData hiding based on compressed VQindices of imagesrdquo Computer Standards and Interfaces vol 31no 6 pp 1143ndash1149 2009
[13] S-C Shie S D Lin and C-M Fang ldquoAdaptive data hidingbased on SMVQpredictionrdquo IEICETransactions on Informationand Systems vol E89-D no 1 pp 358ndash362 2006
[14] C-C Chen and C-C Chang ldquoHigh capacity SMVQ-basedhiding scheme using adaptive indexrdquo Signal Processing vol 90no 7 pp 2141ndash2149 2010
[15] C-H Yang and Y-C Lin ldquoReversible data hiding of a VQ indextable based on referred countsrdquo Journal of Visual Communica-tion and Image Representation vol 20 no 6 pp 399ndash407 2009
[16] Y Linde A Buzo and R M Gray ldquoAlgorithm for vector quan-tizer designrdquo IEEE Transactions on Communications Systemsvol 28 no 1 pp 84ndash95 1980
[17] T Kim ldquoSide match and overlap match vector quantizers forimagesrdquo IEEE Transactions of Image Processing vol 1 no 2 pp170ndash185 1992
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of