A new steganographic scheme based on vector quantisation and search-order coding

8/7/2019 A new steganographic scheme based on vector quantisation and search-order coding

1/6

Published in IET Image ProcessingReceived on 22nd December 2008Revised on 4th March 2009doi: 10.1049/iet-ipr.2008.0251

ISSN 1751-9659

A new steganographic scheme based on vectorquantisation and search-order codingC.-C. Lee W.-H. Ku S.-Y. HuangDepartment of Information and Communication Engineering, Asia University, No. 500, Lioufeng Road, Wufeng Shiang,

Taichung, Taiwan 413E-mail: [email protected]

Abstract: In this paper, the authors propose a new steganographic scheme based on vector quantisation (VQ) andsearch-ordercoding (SOC).The capacity of theproposedscheme is higher than othersbased on VQ andthequality of stego-image is acceptive. In general, the capacity of the schemes based on VQ is usually 1-bit per block. In 2004, Yuet al. proposed a scheme to increase the capacity to 3-bit per pair of blocks. In the proposed scheme, the authorsutilised multiple codebooks andchose thecombinations of per pair of blocks, to embed thesecretdata.The capacitywas increased to 6-bit per pair of blocks, and the quality was identical Yu et al. s scheme.

1 IntroductionComputer and network communication techniques havedeveloped quickly in recent years People attach great importance to secret communication. For this reason,hidden information is being proposed. Hidden informationtechniques are mainly classied into two categories:copyright marking and steganography [1]. Copyright marking such as a signature or a trademark is used toembed copyright information into the host image, thusensuring the copyright. The emphasis on copyright marking is that the watermark hidden in the host imagemust be able to be retrieved, even if the stego-image has

been subjected to lots of manipulation, such as cropping,lossy compression, resizing etc. A watermark is usually very small and composed of visible and imperceptible images.

Steganography is different from watermarking. The goal of steganography is to embed secret data into the host image, insuch a way that the interceptors perceive its presence. In other words, steganography provides a secret communication[2].The key issues in steganography are how to embed moresecret data and yet maintain the stego-images good quality.The technique of data hiding in images can be classiedinto two categories: data hiding in spatial domain schemes

and data hiding in frequency domain schemes. Neither data hiding in spatial domain schemes nor data hiding infrequency domain schemes can be used in other domains.

The most well-known data hiding scheme in the spatialdomain is least-signicant-bits substitution (LSB)[3]. It uses the invisible to the human visual system differencesbetween the host image and stego-image to hide data.There are many well-known schemes based on LSB, suchas optimal LSB, LSB matching, pixel-value differencing multiple-based notational system, and using side match[411]. There are some data hiding schemes using vector quantisation (VQ) compression in the spatial domain,those are different from the schemes mentioned above[1, 1215]. Due to a high compression of VQ, thecapacity for hiding data and the data trafc are much lessthan the mentioned schemes.

In this paper, our proposed scheme is in the spatialdomain, based on VQ and search-order coding (SOC)[14,16]. Prior to this research, the capacity of the schemesbased on VQ was usually 1-bit per block [1, 12, 13]. In2004, Yu et al. [15] proposed a scheme to increase thecapacity to 3-bit per pair of blocks. In this paper, we usemultiple codebooks and choose different combinations of per pair of blocks. The capacity is increased to 6-bit per pair of blocks, and the quality is as good as in Yuet al.sscheme.

This paper is organised as follows. In the next section, weshall briey review VQ and SOC. In Section 3, we shallpropose a new scheme based on VQ and SOC. Next, we

IET Image Process. , 2009, Vol. 3, Iss. 4, pp. 243248 243doi: 10.1049/iet-ipr.2008.0251 & The Institution of Engineering and Technology 2009

www.ietdl.org

Authorized licensed use limited to: Sree Vidyanikethan Engineering College. Downloaded on August 02,2010 at 05:37:22 UTC from IEEE Xplore. Restrictions apply.


2/6

shall compare our scheme with other schemes. Finally, theconclusion will be in Section 5.

2 Preliminaries2.1 VQVQ is a technique which can be used to compress the size of image. As it is a low-bit-rate and lossy compression, VQ canefciently reduce the size of the image.

In Fig. 1 [12], suppose that there exists an image, VQ willdivide the image into many blocks of k * k pixels (usually,k isequal to 4), and strand these pixels into a k2 -dimensionvector. Then we can nd out theN mean vectors (usually N 256) of all vectors creating a codebook, giving each of them an index, and mean vectors. After creating thecodebook, each block is encoded by an index of the closest

codeword in this codebook. Finally, the indices table andcodebook are sent to the receiver. The image can be recoveredfrom the indices table and codebook, but it is different fromthe original image. In other words, the image will be distorted.

2.2 SOC The SOC algorithm which was a compression for VQ indices was proposed by Hsieh and Tsai in 1996[16]. Theauthors found that it was easy to identify the blocks withthe same index around its neighbours in VQ indices table.

Let us describe SOC briey as follows. For example, inFig. 2b [1, 16], if a 2-bit SOC code is used, point (3, 3) is thesearch centre which needs to be encoded, point (3, 2) is thesearch starting point, and the default search path is shown inFig. 2a . After the coding process has started, the searchcentre is set to 53. The search path for the same index startsfrom search point SP (3, 2). Because SP (3, 2) is not markedas 53, it is noted as (00)2 , and moved forward to the next SP.Similarly, SP (2, 3) is marked as (01)2 and the searchcontinued. Finally, SP (2, 1) is found and equal to 53, it isnoted as (10)2 . The search centre is encoded with the SOCcode (10)2 and there will be a 1-bit indicator (0)2 added infront of the SOC code (10)2 . There may exist some repetitive

SP like SP (2, 2), (2, 4) and (3, 1). If their values had beennoted before, they would be neglected. If no SP is equal to53, the search centre will be noted as the original VQ indexvalue (00110101)2 ( 53) and a 1-bit indicator (1)2 will beadded in front of the VQ index value.

3 Proposed schemeOur proposed scheme is used in the spatial domain, and isbased on VQ and SOC. We use multiple codebooks andchoose the combinations of per pair of blocks to embed thesecret data. The embedding procedure and the extracting procedure are described as follows.

3.1 Embedding procedureIn the proposed scheme, we want to apply four codebooks toachieve the goal of embedding 6-bit secret data into pairsof blocks. First, we create a codebook with 64 kind of

Figure 1 VQ for compressing image

Figure 2 Search path and an examplea Search pathb Example of SOC algorithm with a 4 4 index table

244 IET Image Process. , 2009, Vol. 3, Iss. 4, pp. 243248& The Institution of Engineering and Technology 2009 doi: 10.1049/iet-ipr.2008.0251

www.ietdl.org



3/6

codewords by executing a VQ procedure, then make four copies of that codebook and note them as CB1 , CB2 , CB3and CB4 . These codebooks have the same codewords but the indices are different, 0 63 belong to CB1 , 64 127belong to CB2 , 128 191 belong to CB3 and 192 255

belong to CB4 . From these codebooks we then obtain a super codebook with 256 codewords. Now, including theidea of SOC, we will have eight choices for each block,that is, (CB1 , CB2 , CB3 , CB4 ) by using VQ coding and(CB1 , CB2 , CB3 , CB4 ) by using SOC coding. Enlarging

Table 1 Embedding table

Case 6-bits secret data First block Second block Case 6-bits secret data First block Second block

01 (000000) 2 CB1 (VQ) CB1 (VQ) 33 (100000) 2 CB3 (VQ) CB1 (VQ)

02 (000001) 2 CB1 (VQ) CB1 (SOC) 34 (100001) 2 CB3 (VQ) CB1 (SOC)







09 (001000) 2 CB1 (SOC) CB1 (VQ) 41 (101000) 2 CB3 (SOC) CB1 (VQ)

10 (001001) 2 CB1 (SOC) CB1 (SOC) 42 (101001) 2 CB3 (SOC) CB1 (SOC)





















31 (011110) 2 CB2 (SOC) CB4 (VQ) 63 (111110) 2 CB4 (SOC) CB4 (VQ)32 (011111) 2 CB2 (SOC) CB4 (SOC) 64 (111111) 2 CB4 (SOC) CB4 (SOC)


www.ietdl.org



4/6

the idea to pairs of blocks, there will be 64 combinationswhich can be used. Finally, an embedding table for the6-bit secret data is created, shown inTable 1. This tablecan be created as you wish and it must be shared betweenthe sender and the receiver.

The embedding procedure is described as follows:

1. For example, assume that a sender wants to embed(000111)2 in a pair of blocks. According toTable 1, he/she can nd that (000111)2 is in the eighth case, theencoding choices of the rst block and second block areCB1 with VQ and CB4 with SOC, respectively.

2. Search the index of the rst block in CB1 , suppose it is60 (00111100)2 , and add a 1-bit indicator (1)2 in front of the index to indicate it that is a VQ code. (100111100)2is the nal code of the rst block.

3. Search the index of the second block in CB4 , suppose it is250 (11111010)2 , then the sender nds the SOC code,that is (10)2 , by executing SOC search. Add a 1-bit indicator (0)2 in front of the index to indicate that it is a SOC code. (010)2 is the nal code of the second block.

4. Execute (1), (2) and (3) until all the secret data areembedded.

5. Send the super codebook and all the codes to the receiver.

3.2 Extracting procedureNote that the sender and the receiver share the embedding table and the super codebook is transmitted with thosecodes after the embedding procedure. To continue theexample in the embeddiing procedure, the receiver will ndthat the indicator is (1)2 in the rst block when the receiver wants to extract the secret data. It means that this is a VQ code, and the code is 60 (00111100)2 . The receiver canlook up the super codebook and nd out that it is an indexin CB1 . Note that the rst block is coded by CB1 with

VQ. Then the receiver will nd that the indicator is (0)2 inthe second block, it means this is a SOC code, and thecode is (10)2 . The receiver can nd that the VQ code is250 (11111010)2 by executing a SOC search. Thereceiver knows that it is an index in CB4 after looking upthe super codebook. The receiver obtain the rst block coded by CB1 with VQ and the second block coded by CB4 with SOC; then the receiver will nd that the secret data are (000111)2 in this pair of blocks by looking for theembedding table.

4 ComparisonAs we know, the key issues in steganography are how toembed more secret data and still maintain the stego-images good quality. The comparisons of embedding capacity and image quality of Chang et al.s [1] scheme, Yuet al.s [15] scheme and our proposed scheme are as follows.

To compare the performance, the six gray-scale 512 512host images are toys, couple, boats, barbara, baboon andpeppers, with the results shown inTable 2 [16]. According to the experimental results, we can nd that the embedding capacity of the proposed scheme is twice as good as Yuet al.s scheme. The average capacity of our proposedscheme is 49 152 bits, whereas Yuet al.s scheme is 24 576only.

The common way to compare the quality of two images isto compute the peak signal to noise ratio (PSNR), the

formula is as follows:

PSNR 10log 102552

MSE(dB)

where

MSE 1

M N XM 1

i 0 XN 1

j 0(P (i , j ) P 0(i , j ))2

Table 2 Comparisons of embedding capacity

Images

[1] [15] The proposed scheme

EmbeddedBlocks

Embedded data(bits)

Embeddedblocks

Embedded data(bits)

Embeddedblocks

Embedded data(bits)

toys 16 384 16 384 16 384 24 576 16 384 49 152

couple 16 384 16 384 16 384 24 576 16 384 49 152

boats 16 384 16 384 16 384 24 576 16 384 49 152

barbara 16 384 16 384 16 384 24 576 16 384 49 152

baboon 16 384 16 384 16 384 24 576 16 384 49 152

peppers 16 384 16 384 16 384 24 576 16 384 49 152average 16 384 16 384 16 384 24 576 16 384 49 152


www.ietdl.org



5/6

andM

is the length of image,N

is the width of image,P

(i ,

j )is the value of the host image in pixel (i , j ) and P 0(i , j ) is the

value of the stego-image in pixel (i , j ) [2]. The larger thevalue of PSNR, the better the image quality will be. Thecomparisons of image quality are shown inTable 3 [16].The results show that the image quality of the proposedscheme is the same as Yuet al.s scheme, becauseSOC is an undistorted coding, the image quality will not be affected.

5 Conclusions

A high capacity data hiding scheme for VQ compressedimage is presented in this paper. The proposed scheme canembed 6-bit secret data for per pair for VQ blocks, and theimage quality is the same as the Yuet al.s schemes. Theproposed scheme can fend off the image compressionattack, which is a common problem in those data hiding schemes used in the spatial domain.

6 AcknowledgmentThis research was partially supported by the National ScienceCouncil, Taiwan, Republic of China under contract no.:

NSC97-2218-E-468-010.

7 References

[1] CHANG C.C., CHEN G.M., LIN M.H.: Information hiding basedon search-order coding for VQ indices, Pattern Recognit.Lett. , 2004, 25 , pp. 12531261

[2] WU N.I., HWANG M.S.: Data hiding: current status andkey issues, Int. J. Netw. Security (IJNS) , 2007, 4, (1), pp. 19

[3] BENDER W., GRUHL D., MORIMOTO N., LU A.: Techniquesfor data hiding, IBM Syst . J. , 1996, 35 , (3 4 ),pp. 313336

[4] CHAN C.K., CHENG L.M.: Hiding data in images bysimple LSB substitution, Pattern Recognit. , 2004, 37 , (3),pp. 469474

[5] CHANG C.C., HSIAO J.Y., CHAN C.S.: Finding optimal least-signicant-bit substitution in image hiding by dynamicprogramming strategy, Pattern Recognit. , 2003., 36 , (7),pp. 15831595

[6] CHANG C.C., TSENG H.W.: A steganographic method fordigital images using side-match, Pattern Recognit. Lett. ,2004, 25 , (12), pp. 14311437

[7] K ER A. D.: Steganalys is o f LSB match ing ingrayscale images, IEEE Signal Process. Lett. , 2005, 12 ,pp. 441444

[8] MIELIKAINEN J.: LSB matching revisited, IEEE Signal Process. Lett. , 2006, 13 , pp. 285287

[9] WANG R.Z., LIN C.F., LIN J.C.: Image hiding by optimal LSBsubstitution and genetic algorithm, Pattern Recognit. ,2001, 34 , (3), pp. 671683

[10] WU D.C., TSAI W.H.: A steganographic method for images

by pixel-value differencing, Pattern Recognit. Lett. , 2003,24 , (9 10), pp. 16131626

[11] ZHANG X., WANG S.: Steganography usingmul tiple -base notational system and humanvision sensitivity, IEEE Signal Process. Lett. , 2005, 12 ,pp. 6770

[12] DU W.C., HSU W.J.: Adaptive data hiding based on VQ compressed images, IEE Proc. Vision Image Signal Process. , 2003, 150 , pp. 233238

[13] JO M., KIM H.D.: A digital image watermarking schemebased on vector quantisation, IEICE Trans. Inf. Syst. , 2002,E85-D, (6), pp. 10541056

Table 3 Comparisons of image quality

ImagesVQ [1] [15] Proposed scheme

PSNR PSNR Distortion PSNR Distortion PSNR Distortion

toys 28.104 28.104 0 27.627 0.476 27.627 0.476

couple 27.269 27.269 0 26.826 0.443 26.826 0.443

boats 27.901 27.901 0 27.582 0.319 27.582 0.319

barbara 25.380 25.380 0 25.063 0.317 25.063 0.317

baboon 24.252 24.252 0 23.987 0.265 23.987 0.265

peppers 28.775 28.775 0 28.303 0.471 28.303 0.471

average 26.947 26.947 0 26.565 0.382 26.565 0.382


www.ietdl.org



6/6

[14] LINDE Y., BUZO A., GRAY R.M.: An algorithm for vectorquantiser design, IEEE Trans. Commun. , 1980, 28 , (1),pp. 84 95

[15] YU Y.H., CHANG C.C., HU Y.C.: A steganographic method forhiding data in VQ encoded images. Proc. 2004 Int. Symp.

Intelligent Multimedia, Video and Speech Processing,Hong Kong, August 2004, pp. 358361

[16] HSIEH C.H., TSAI J.C.: Lossless compression of VQ indexwith search-order coding, IEEE Trans. Image Process. ,1996, 5, pp. 15791582


www.ietdl.org

A new steganographic scheme based on vector quantisation and search-order coding

Documents

Transcript of A new steganographic scheme based on vector quantisation and search-order coding