Research Article Perceptual Hashing-Based Robust Image...

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2013 Article ID 791814 9 pageshttpdxdoiorg1011552013791814

Research ArticlePerceptual Hashing-Based Robust Image Authentication Schemefor Wireless Multimedia Sensor Networks

Hongxia Wang and Bangxu Yin

School of Information Science and Technology Southwest Jiaotong University Chengdu 610031 China

Correspondence should be addressed to Hongxia Wang hxwanghomeswjtueducn

Received 29 January 2013 Accepted 13 March 2013

Academic Editor Muhammad Khurram Khan

Copyright copy 2013 H Wang and B YinThis 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

Image authentication is critical for secure image transmission and storage in a wireless multimedia sensor network (WMSN) Inthis paper we propose a perceptual hashing-based robust image authentication scheme which applies the distributed processingstrategy for perceptual image hashes and can provide compactness visual fragility perceptual robustness and security in digitalimage authentication for WMSN In the proposed scheme first the cluster head node generates a secure pseudorandom chaoticsequence with keys and sends it to the image capturing node then the image capturing node uses the received chaotic sequence todivide randomly the captured image into several overlapping rectangles after that two gravity centers of each rectangular regionblock are calculated and finally the binary distance of the two gravity centers will be calculated in each general cluster membernodeThe cluster head node receives the binary sequence of the distance from all of the general clustermember nodes and combinesthem to form the perceptual hashing sequence to be sent to the base station for image authentication purpose Experimental resultsshow that the proposed scheme has satisfactory authentication ability and can ensure not only the visual fragility for perceptuallydistinct images but also robustness for perceptually identical images via image rotation JPEG compression and noise burring

1 Introduction

Wireless sensor networks (WSNs) are going to be widelyused in the near future due to their breadth of applicationsby military exploration teams researchers and so on Mostof this research is concerned with scalar sensor networksthatmeasure physical phenomena such as temperature lighthumidity pressure acoustic sensor or location of objectsthat can be conveyed through low-bandwidth and delay-tolerant data streams Recently the focus is shifting towardresearch aimed at revisiting the sensor network paradigmto enable delivery of multimedia information such as amonitoring data digital image audio and video streams aswell as scalar data [1] This effort will result in distributednetworked systems referred to in this paper as wirelessmultimedia sensor networks (WMSNs) The WMSNs willenable new applications such as multimedia surveillancetraffic enforcement and control systems advanced health caredelivery structural health monitoring and industrial processcontrol Consequently it will bring new security of challengesas well as new opportunities Secure and robust multimedia

communications become increasingly important for energy-constrained WMSNs [2] As one of the security techniquesimage authentication is critical for secure image transmissionand storage in WMSNs However conventional data authen-tication schemes cannot be applied directly to WMSN due tothe constraints on limited energy and computing resourcesin sensor nodes Those constraints pose great challenges toWMSN development and motivate us to design a properauthentication strategy for WMSNs

For open communication channel WSNs are vulnerableto various attacks and the security in WSNs is required Asecure hierarchical key management scheme in WSNs wasproposed in [3] The security analysis and simulation showthe scheme can prevent several attacks effectively and reducethe energy consumption In WSNs communications Han etal [4] described six types of attacks including communicationattacks attacks against privacy sensor node targeted attackspower consumption attacks policy attacks and cryptologyattacks on key management In communication attackseavesdropper can easily access or even manipulate messagesuch as injecting cropping and tampering So the receiver

2 International Journal of Distributed Sensor Networks

needs to make sure that the data used in any decision-making process originates from the correct source Dataauthentication prevents unauthorized parties from partici-pating in the networks and legitimate nodes should be ableto detect messages from authorized nodes and reject themIn [5] a robust user authentication scheme for WSNs wasproposed The scheme takes an advantage of the two-factorauthentication concept to provide a secure authenticationsystem offering balanced features in terms of security andperformance

The authenticity of data and commands is also a crit-ical requirement for the correct behavior of a WMSN[6] Digital images are becoming widely used in WMSNsas a kind of common multimedia information Thereforethe key establishment technique should guarantee that theimage communication and storage have a way for ver-ifying the authenticity creditability and integrity of thereceived image in WMSNs However resource constraintsin sensor networks (such as limited battery power andbandwidthcomputation capability) pose challenges for theimage authentication technique Conventional binary dataauthentication schemes could provide data integrity in astrict sense regardless of multimedia content However thoseschemes are not applicable toWMSNsbecause only simple biterrors during data transmission can lead to the authenticationfailing in spite of preserving multimedia content [7 8] Onthe other hand watermarking-based image content authenti-cation techniques are robust against bit errors packet lossesand compression distortionHowever watermark embeddingcreates extra source coding overheads and complicates trans-mission protocol design in WMSNs [9] which donrsquot adaptwell to WMSNs due to the constraints on limited energy andcomputing resources

In [10] an optimized content-aware authenticationscheme for JPEG 2000 images over lossy channels wasproposed An acyclic authentication graph was developed tooptimize the trade-off between the expected image distortionand the cryptohash tagging cost through the computationbased on packet loss probability and visual importance levelof the image packets The work reported in [11] proposeda JPEG 2000 compatible stream authentication scheme thatsignificantly reduced the computational complexity and hadonly a minimal authentication dependency overhead inWMSNs Moreover an authentication-aware wireless net-work resource allocation scheme was developed to reduceimage distortion and energy consumption during transmis-sion The scheme significantly improved the authenticatedimage quality even under strict communication energy con-sumption constraints in WMSN In [12] a rate-distortionoptimization authentication scheme for H264 video trans-missions was proposed A video packet transmission sched-uler was designed tominimize the visual distortion under thelimitation of total bit budget and authentication dependencyAnother related work regarding bite rrors robust imageor video authentication was given in [13ndash15] However allof these approaches are not able to be applied directly toWMSNs due to the energy constraint In this paper wepropose a perceptual hashing-based robust authenticationscheme for WMSNs Based on the distributed processing

strategy for perceptual image hashes the proposed schemecan provide compactness visual fragility perceptual robust-ness and security for image authentication in WMSNs

2 Perceptual Image Hashing Extraction

A perceptual image hashing function maps an image to ashort binary string as a digest based on an imagersquos appearanceto the human eye Perceptual image hashing is a class of one-way mappings from image presentations to a perceptual hashvalue in terms of their perceptual content Given an image 119868

and its perceptually similar copywithminor distortion 119868119889 theimage hashing function119867119896(sdot) depends on the secret key 119896 In[16] the desirable properties of perceptual hashing function119867119896(sdot) can be summarized as follows

(i) One-Way Function Ideally the hash generationshould be noninvertible

119868 997891997888rarr 119867119896 (119868) (1)

(ii) CompactnessThe size of the perceptual hashing valueshould be much smaller than that of the originalimage 119868

Size (119867119896 (119868)) ≪ Size (119868) (2)

(iii) Perceptual Robustness Perceptually identical imagesshould have similar perceptual hashing values

Pr 119867119896 (119868) asymp 119867119896 (119868119889) ge 1 minus 120576 0 le 120576 lt 1 (3)

(iv) Visual Fragility Perceptually distinct images shouldhave different perceptual hashing values

Pr 119867119896 (119868) =119867119896 (1198681015840) ge 1 minus 120591 0 le 120591 lt 1 (4)

(v) Security The perceptual hashing is intractable with-out the secret key

Pr 119867119896 (119868) =1198671198961015840 (119868) ge 1 minus 120575 0 le 120575 lt 1 (5)

All of the above parameters 120576 120591 and 120575 should be close to zeroBased on THE above properties perceptual image hash-

ing can be usually applied to image content identificationimage indexing content authentication and so forth Inparticular a perceptual hash function should have a propertythat is two images that look the same map to the samehash value even if the images have small bit-level differencesThis differentiates a perceptual hash from traditional cryp-tographic hashes such as SHA-l and MD5 In cryptographythe hash function is typically used for digital signature toauthenticate the message being sent so that the receivercan verify its source A key feature of conventional hashingalgorithms such as SHA-l andMD5 is that they are extremelysensitive to the input data that is changing even one bit ofthe input message will change the output dramatically How-ever image data often undergoes various content-preservingmanipulations such as lossy compression channel additive

International Journal of Distributed Sensor Networks 3

noise image enhancement scaling bit errors and packetlosses during wireless transmission and storage in WMSNsThese distortions are usually insignificant and image hashesshould be robust to unmalicious distortions On the contrarysome malicious manipulations could introduce perceptuallysignificant distortions for example object insertion removaland substitution and it is desirable that the image hash issensitive to perceptually significant attacks Therefore imagehashes should be robust to unmalicious distortions but sensi-tive to malicious manipulations for the image authenticationpurpose [17]

21 Image Random Blocking by Chaotic Sequence In order toenhance the security of the perceptual hashing algorithm weuse a secure pseudorandom sequence with the key to dividerandomly the digital image into 119862 overlapping rectanglesand the key controls the number of rectangles and thepseudorandom sequence The image blocking can also makeup the disadvantage that the extracted image features can onlydescribe global characteristics of an image

As a phenomenon found in a nonlinear dynamic systemchaos is deterministic and random-like Based upon thesensitive dependence of chaotic systems on their initial con-ditions a large number of nonperiodic continuous broad-band frequency spectrum noise-like yet deterministic andreproducible signals can be generated So chaos is very usefulfor generating secure pseudorandom sequences

The chaotic maps (6) and (7) are given by

119885119899+1 = 119906119885119899 (1 minus 119885119899) 119899 = 1 2 (6)

119878119899+1=(1 + 03 (119878119899minus1 minus 108) + 3791198782

119899+ 1001 times 119885

2

119899)mod 3

(7)

where 357 lt 119906 le 4 is the chaotic system parameter and 0 lt

1198851 lt 1 and minus15 lt 1198780 1198781 lt 15 are the initial values of thetwo chaotic systems 119885119899 in formula (7) is generated by (6)When 119906 = 39 we compute the value space of 1198851 as follows

Suppose 1198851 = 0 lt 1198851(119894) lt 1 | 119894 = 1 2 1198711 and1198711 is an integer which is large enough and generates chaoticsequences 119885 = 119885(119894 119895) | 119894 = 1 2 1198711 119895 = 1 2 1198712where 1198711 represents the number of chaotic sequences and 1198712

means the length of each chaotic sequence When 1198851015840

1= 0 lt

1198851(119894) + 119889 lt 1 | 119894 = 1 2 1198711 and 119889 is close to zero anothergroup of chaotic sequence 1198851015840 = 119885

1015840(119894 119895) | 119894 = 1 2 1198711 119895 =

1 2 1198712 is generated We use function 119910 = Ω(119889) to testthe value space of 1198851 as follows

119910 = Ω (119889) =

[sum1198711

119894=1sum1198712

119895=1

10038161003816100381610038161003816119885 (119894 119895) minus 119885

1015840(119894 119895)

10038161003816100381610038161003816]

(1198711 times 1198712) (8)

The curve of function119910 = Ω(119889) is shown in Figure 1 FromFigure 1 it is seen that when 119889 = 10

minus19 119910 asymp 0 So the value

space of 1198851 is 1019 Similarly the value spaces of 119906 1198780 and 1198781

are shown in Figures 1 and 2 They are 1 times 1015 3 times 10

14 and3 times 10

16 respectivelyTherefore when the difference of initial values or chaotic

parameters of chaotic system is greater than some specific

035

03

025

02

015

01

005

00 10

minus510minus10

10minus15

10minus20

10minus25

119910

119889

1198851119906

Figure 1 The value spaces of 1198851 and 119906

14

12

1

08

06

04

02

0

11987811198780

0 10minus5

10minus10

10minus15

10minus20

10minus25

119910

119889

Figure 2 Value spaces of 1198781 and 1198780

value two different chaotic sequences will be generated For1198851 the difference should be greater than 1019 and similarlyfor others Let 119906 in (6) and 1198780 as secret keys be denoted by 1198961

and 1198962 respectively Consequently we will use the generatedchaotic sequences with keys 1198961 and 1198962 to divide randomlythe digital image into 119862 overlapping rectangles for securitypurpose

According to the image size we adaptively select properbits of the chaotic sequence as the coordinate of 119909-axis and119910-axis length and width of the random region to preventout boundary denoted by a quaternion 119862 (119862119909 119862119910 lengthwidth) So the image is divided into 119862 overlapping rectan-gular regions shown in Figure 3 Because the quaternion israndomly generated the rectangular areas are random


Figure 3 Image divided randomly into overlapping blocks

22 Robust Local Feature Points Based on Gravity Center ofRandom Blocks The local features of the image should benot only stable under geometric transforms such as rotationand scaling but also robust to insignificant distortions suchas additive noising and blurring bit errors and packet lossesduring transmission inWMSNs Based on the image randomblocking in Section 21 we propose a robust local featurepoints extraction method using the gravity center of randomblocks The extracted robust local features will then be usedto generate perceptual hashes in Section 23

The two-dimensional (2D) moment can be directly usedin the interested regions and does not need to separate themfrom the whole image High-order moments are more sensi-tive to noise while the low-order moments are insensitive tonoise and bit errors which is beneficial to the characterizationof collectivity for the regions

The 2D (119901 + 119902)th order origin moment of a continuousimage 119868 is defined as [18]

119872119901119902 = int119868

119909119901119910119902119891 (119909 119910) 119889119909 119889119910 119901 119902 = 0 1 2 (9)

where 119891(119909 119910) represents the gray-level value at location(119909 119910) For digital image the integrals are replaced by sum-mations and119872119901119902 becomes

119872119901119902 = sum

(119909119910)isin119868

119909119901119910119902119891 (119909 119910) 119901 119902 = 0 1 2 (10)

The gravity center 119866 (119898119909 119898119910) of the image is defined interms of the zero-order moment and first-order moments asfollows

119898119909 =11987210

11987200

119898119910 =11987201

11987200

(11)

where the zero-order moment11987200 represents the area of theimage clearly

In order to obtain the local feature the coordinate of thegravity center 119866119896(119898

119896

119909 119898119896

119910) of each random rectangular block

is calculated as

119898119896

119909=

sum119909sum119910119909 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119898119896

119910=

sum119909sum119910119910 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119896 = 1 2 119862

(12)

where 119891119896(119909 119910) represents the gray-level value at location

(119909 119910) in the 119896th random rectangular block and 119862 is thenumber of random rectangular blocks Thus there are total119862 gravity centers of the pseudorandom rectangular blocksdenoted by 119866 = 1198661 1198662 119866119896 119866119862 The local featureinformation of the image can be obtained by calculation ofeach blockrsquos gravity center which improves the ability todistinguish different images

The gravity center of the image has geometrically invari-ant property In this paper the rotation invariability isanalyzed as an example After a rotation by an angle 120579 aboutthe original the first-order moments are given by

119872119903

10= sum

(119909119910)isinΩ

(119909 cos 120579 + 119910 sin 120579) 119891 (119909 119910)

= 11987210 cos 120579 + 11987201 sin 120579

119872119903

01= sum

(119909119910)isinΩ

(119910 cos 120579 minus 119909 sin 120579) 119891 (119909 119910)

= 11987201 cos 120579 minus 11987210 sin 120579

(13)

and the zero-order moment 11987211990300

= 11987200 So after a rotationby an angle 120579 the changed gravity center is

119866119903

119909=

119872119903

10

119872119903

00

= 119866119909 cos 120579 + 119866119910 sin 120579

119866119903

119910=

119872119903

01

119872119903

00

= 119866119910 cos 120579 minus 119866119909 sin 120579

(14)

Namely

[

[

119866119903

119909

119866119903

119910

]

]

= [cos 120579 sin 120579

minus sin 120579 cos 120579] [119866119909

119866119910] (15)

Thus the rotation invariability is held In similar analysisother geometrical invariability characteristics can also beobtained

Figure 4 shows the robustness of gravity centers undergeometric distortions and common image processing Notethat times denotes the virtual locations gravity centers in thedistorted image and o denotes theoretical locations of gravitycentersOnce the two symbols are coincident the geometricalinvariability of gravity centers and the strong robustness toadditive noise and JPEG compression will be indicated


650

600

550

500

450

400

350

300

250

200100 200 300 400 500 600 700

119898119909

119898119910

(a) Rotate 30∘

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

119898119909

119898119910

(b) Scale 2 times

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910

(c) Additive white Gaussian noise (variance is 46)

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910

(d) JPEG compression (quality factor is 50)

Figure 4 Robustness of gravity centers under geometric distortions and common image processing

23 Perceptual Image Hashes Generation The supplementimage block of each random rectangular block is defined as

(119909 119910) = 119877level minus 119891 (119909 119910) (16)

where 119877level is the maximum gray-scale level of the imageLikewise we obtain the gravity center of the supplementimage block For each rectangular block we calculate itssupplement image blockrsquos gravity center and obtain total 119862gravity centers of the supplement image blocks The normalgravity center of image usually lies around the center ofimage so does that of the supplementary image Thus thedistance between the two gravity centers of image and itssupplement image is short We devise a solution by making amodification of the gravity center The improved supplement

image blocksrsquo gravity center 119896

(119909 119910) of 119896th rectangularimage block is obtained by

119896

119909=

sum119909sum119910119909 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

119896

119910=

sum119909sum119910119910 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

(17)

where Δ 1 gt 0 is the quantification step Through suchmodification we enlarge the distance between the two gravitycenters on one hand On the other hand the parameterΔ 1 guarantees the robustness against malicious attacks incalculating the two gravity centers

In order to generate the perceptual image hashes wecalculate the distance of the two gravity centers betweeneach rectangular block and its supplement image block Con-sidering the constraints on limited energy and computing


BS CH

1198751

1198752

1198753

1198754

119875119899

119878

middot middot middot

Figure 5 Structure of clustering in WMSNs

resources inWMSNs we use 119871infin norm tomeasure the spatialdistance 119863119896 between the locations of the two gravity centersfor each rectangular block as follows

119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)

Let Δ 2 gt 0 be another quantification step and the distance119863119896 between the two gravity centers can be quantified as

119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)

where lfloorsdotrfloor denotes the floor function Obviously the quan-tified distance 119863119896 will be decimal integer Moreover thedistance119863119896 ismore robust against common image processingand the bit errors during transmission

Then we convert the decimal 119863119896 to binary sequence(1198871198961 1198871198962 119887119896119895)2

119887119896119895 isin 0 1 and finally combine thebinary sequences of each distance together to form theperceptual hashing sequence119867 as follows

119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

3 Distributed Processing Strategy forPerceptual Image Hashes

In WSNs clustering expedites many desirable functions andprovidesmany advantages such as load balancing energy sav-ings and distributed key management The most prominentbenefit of clustering is that it can greatly reduce the energyconsumption of nodes and lengthen the network lifetime [19]In this paper in order to adapt well to the limited powerresources and computational capabilities in sensor nodes weconsider clustering-based WMSNs with densely distributednodesThe structure of a clustering is shown in Figure 5 Eachclustering consists of a cluster head (CH) several generalcluster member nodes and a camera sensor that captures thedigital image In Figure 5 BS represents the base station CHis a cluster head node 119878 is the image capturing node and1198751 sim 119875119899 are the general cluster member nodes whose eachnode is assigned a fixed ID

The distributed processing strategy is as follows

Step 1 The cluster head node CH generates a secure pseudo-random chaotic sequence with the keys 1198961 and 1198962 accordingto the method in Section 21 and sends this chaotic sequenceto the image capturing node 119878 In addition the ID of eachgeneral cluster member node 119875119894 (119894 = 1 2 119899) is also sentto node 119878

Step 2 When the image capturing node 119878 captures an imageit will use the received chaotic sequence to divide randomlythe captured image into 119862 overlapping rectangles Moreovereach rectangle block is sent to the general cluster membernode 119875119894 Note that the chaotic sequence is mapping to the IDof the general cluster member nodes one by one

Step 3 The distance of the two gravity centers for eachrectangle block will be calculated in each general clustermember node 119875119894 and its corresponding binary sequence(1198871198961 1198871198962 119887119896119895)2

can be generated by the method in Sec-tion 23 which is sent to the cluster head node CH

Step 4 The cluster head node CH receives the binary seq-uences (1198871198961 1198871198962 119887119896119895)2

119896 = 1 2 119862 from all of thegeneral cluster member nodes then combines them to formthe perceptual hashing sequence and finally the cluster headnode CH sends the perceptual hashing sequence to the basestation for image authentication purpose

4 Perceptual Hashing-Based ImageAuthentication

When we identify the received image firstly the perceptualhashing sequence is obtained from the base station and ismatched with the perceptual hashing sequence generated bythe dubitable image If the Hamming distance between twoperceptual hashing sequences is less than the specified thre-shold the image will be deemed an authentic versionOtherwise the image is forged

The image authentication framework is shown in Fig-ure 6 The steps are as follows

Step 1 The perceptual hashing sequence 1198671 is received fromthe base station

Step 2 The dubitable image is partitioned into 119862 randomrectangular blocks by the same secret keys 1198961 and 1198962 like per-ceptual hashing generation process described in Section 21then the distance of the two gravity centers for each rectangleblock is calculated and quantified after that the robust featureis extracted Thus another perceptual hashing sequence 1198672

can be restructured like the description in Section 23

Step 3 Setting a threshold 119879 gt 0 normalized Hamming dis-tance is calculated by

119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962

Image to beauthenticated

Robust featureextraction

Restructuredperceptual hash1198672

Receivedperceptual hash1198671

Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879

Figure 6 Image authentication process

0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01

Normalized Hamming distance

Appe

aran

ce ti

me

Figure 7 Visual fragility test

where 119871 is the length of perceptual Hashing sequence If119863119867(1198671 1198672) le 119879 the image will be deemed an authenticversion Otherwise if 119863119867(1198671 1198672) gt 119879 the image will beforgedThe smaller the normalized Hamming distance is thestronger the robustness is Ideally the normalized Hammingdistance for the perceptually identical images is close to 0while the normalized Hamming distance of two differentimages is close to 05

5 Experimental Results and Analysis

51 Visual Fragility of Perceptual Hashes The visual fragilityrepresents that perceptually distinct images generate differentperceptual hashes We randomly select 80 images sized 300 times

300 The parameters are set as follows 1198961 = 39 1198962 =

12 Δ 1 = 10 Δ 2 = 4 and the number of random rectangleblocks is 150 Then we calculate their perceptual hashesand the Hamming distance between two perceptual hashingvalues 119863119867(1198671 1198672) Finally 6320 matching results can beobtained The statistical histogram distribution is shown in

0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce

Figure 8 Robustness to image rotation

Figure 7 FromFigure 7 we see the results can be approximateto the Gaussian distribution with the expectation 120583 = 02924

and standard deviation 120590 = 00574 Setting the threshold119879 = 0125 the conflict probability will be

119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)

As a result the conflict probability is very small Hencethe proposed perceptual hashing can ensure the visual fragil-ity

52 Perceptual Robustness To test the perceptual robustnessto geometric transforms we rotate the ldquoLenardquo image sized512 times 512 with different degrees and calculate the perceptualhashes Compare the perceptual hashing value of rotatedimage with the originalThe relationship of the rotation angleand the normalized Hamming distance is shown in Figure 8When the image rotation angle is within 5∘ the normalizedHammingdistance is less than 03 so the proposed perceptualhashing algorithm is robust to image rotation within 5∘


30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor

Figure 9 Robustness to image compression

10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce

Multiplicative random noiseAdditive Gaussian white noise

Figure 10 Robustness to noise blurring

Figure 9 shows the robustness to image compression Whenthe quality factor of JPEG compression is changed from 90 to30 the normalized Hamming distance between the originalimage and the compressed image is less than 03 so theproposed perceptual hashing algorithm is robust to imagecompression The smaller the quality factor is the largerthe compression degree is From Figure 9 we see that therobustness is gettingmore andmore strongwith the accretionof the quality factor

Figure 10 shows the robustness to additive Gaussianwhite noise and uniformly distributed multiplicative randomnoise When the image is blurred by different noise degreeswith variance 6554 sim 6554 we calculate the perceptualhashes From Figure 10 we see that when the variance is lessthan 6554 the normalized Hamming distance between theoriginal image and the noise blurred image is less than 03That is to say the proposed algorithm is robust to Gaussian

Table 1 Perceptual hashing segments by different keys

Segment 1 (1198961= 39 119896

2= 12) sdot sdot sdot100101110001101100001101000sdot sdot sdot

Segment 2 (1198961 = 391 1198962 = 12) sdot sdot sdot100011011001010010011000100sdot sdot sdotSegment 3 (1198961 = 39 1198962 = 121) sdot sdot sdot101010011000100111001001110sdot sdot sdot

white noise and uniformly distributed randommultiplicativenoise during image transmission and storage

53 Security Because the chaotic sequence is nonperiodicand sensitive to the initial value the chaotic maps (6) and(7) are used to generate the pseudorandom sequence inthis paper Then the digital image is randomly dividedinto overlapping rectangular regions by the chaotic sequencefor perceptual image hashing extraction Thus if chaoticinitial values are changed that is the keys are different theextracted perceptual image hashing will be different Table 1lists the segments of the generated perceptual image hashesby different keys which indicates the perceptual hashingis intractable without the secret key We can calculate thatthe normalized Hamming distance is 04167 between thecorresponding locations of the perceptual hashing segment1 and segment 2 and that it is 04283 between segment 1 andsegment 3 The two normalized Hamming distances are allclose to 05 so the securitymeets the application requirementTherefore without knowing the key even if the perceptualimage hashing algorithm is known the correct perceptualhashing value generated by the image cannot be leaked

6 Conclusions

In this paper we have proposed a robust image authenticationmethodology for authenticity creditability and integritytransmission and storage of authenticated images based onthe perceptual hashing technique in WMSNs First a gravitycenter-based perceptual image hashing algorithm is proposedwith compactness perceptual robustness visual fragility andsecurity The generated perceptual hashing value is a shortbinary string as a digest of the image in order to tacklethe problem of severe energy constraints and perceptualimage redundancy in WMSNs Furthermore a distributedprocessing strategy for perceptual image hashes is developedto meet the limited computing resources of sensor nodesin WMSNs The experimental results demonstrate that ourscheme has the satisfactory authentication performance forperceptually distinct and identical images

Acknowledgments

This research was supported by the National Natural ScienceFoundation of China (NSFC) under Grant no 61170226 theFundamental Research Funds for the Central Universitiesunder Grants nos SWJTU11CX047 SWJTU12ZT02 theYoung Innovative Research Team of Sichuan Province underGrant no 2011JTD0007 and Chengdu Science and Technol-ogy program under Grant no 12DXYB214JH-002


References

[1] I F Akyildiz T Melodia and K R Chowdury ldquoWirelessmultimedia sensor networks a surveyrdquo IEEE Wireless Commu-nications vol 14 no 6 pp 32ndash39 2007

[2] H Wang M Hempel D Peng W Wang H Sharif and HH Chen ldquoIndex-based selective audio encryption for wirelessmultimedia sensor networksrdquo IEEE Transactions on Multime-dia vol 12 no 3 pp 215ndash223 2010

[3] Y Y Zhang X Z Li J M Liu J C Yang and B J Cui ldquoAsecure hierarchical key management scheme in wireless sensornetworkrdquo International Journal of Distributed Sensor Networksvol 2012 Article ID 547471 8 pages 2012

[4] S Han E Chang L Gao and D Tharam ldquoTaxonomy ofattacks on wireless sensor networksrdquo in Proceedings of the 1stEuropean Conference on Computer Network Defense pp 97ndash105Glamorgan UK December 2005

[5] S G Yoo K Y Park and J Kim ldquoA security-performance-bal-anced user authentication scheme forwireless sensor networksrdquoInternational Journal of Distributed Sensor Networks vol 2012Article ID 382810 11 pages 2012

[6] B Harjito and S Han ldquoWireless multimedia sensor networksapplications and security challengesrdquo in Proceedings of the 5thInternational Conference on Broadband Wireless ComputingCommunication and Applications (BWCCA rsquo10) pp 842ndash846November 2010

[7] M K Khan ldquoFingerprint biometric-based self-authenticationand deniable authentication schemes for the electronic worldrdquoIETE Technical Review vol 26 no 3 pp 191ndash195 2009

[8] Z Li Q Sun Y Lian and C W Chen ldquoJoint source-channel-authentication resource allocation and unequal authenticityprotection for multimedia over wireless networksrdquo IEEE Trans-actions on Multimedia vol 9 no 4 pp 837ndash850 2007

[9] W Wang D Peng H Wang H Sharif and H H ChenldquoEnergy-distortion-authentication optimized resource alloca-tion for secure wireless image streamingrdquo in Proceedings of theIEEE Conference on Wireless Communications and Networking(WCNC rsquo08) pp 2810ndash2815 April 2008

[10] Z Zhang Q Sun W C Wong J Apostolopoulos and SWee ldquoAn optimized content-aware authentication scheme forstreaming JPEG-2000 images over lossy networksrdquo IEEE Trans-actions on Multimedia vol 9 no 2 pp 320ndash331 2007

[11] W Wang D Peng H Wang H Sharif and H H ChenldquoA multimedia quality-driven network resource managementarchitecture for wireless sensor networks with stream authen-ticationrdquo IEEE Transactions on Multimedia vol 12 no 5 pp439ndash447 2010

[12] Z Zhang Q Sun W C Wong J Apostolopoulos and SWee ldquoRate-distortion-authentication optimized streaming ofauthenticated videordquo IEEE Transactions on Circuits and Systemsfor Video Technology vol 17 no 5 pp 544ndash557 2007

[13] Q Sun and S F Chang ldquoA secure and robust digital signaturescheme for JPEG2000 image authenticationrdquo IEEE Transactionson Multimedia vol 7 no 3 pp 480ndash494 2005

[14] Q Sun D He and Q Tian ldquoA secure and robust authenticationscheme for video transcodingrdquo IEEE Transactions on Circuitsand Systems for Video Technology vol 16 no 10 pp 1232ndash12442006

[15] D Skraparlis ldquoDesign of an efficient authentication methodfor modern image and videordquo IEEE Transactions on ConsumerElectronics vol 49 no 2 pp 417ndash426 2003

[16] X D Lv and Z J Wang ldquoPerceptual image hashing based onshape contexts and local feature pointsrdquo IEEE Transactions onInformation Forensics and Security vol 7 no 3 pp 1081ndash10932012

[17] M Tagliasacchi G Valenzise and S Tubaro ldquoHash-basedidentification of sparse image tamperingrdquo IEEE Transactions onImage Processing vol 18 no 11 pp 2491ndash2504 2009

[18] K R Castleman Digital Image Processing Prentice Hall NewYork NY USA 1998

[19] G Wang D Kim and G Cho ldquoA secure cluster formationscheme in wireless sensor networksrdquo International Journal ofDistributed Sensor Networks vol 2012 Article ID 301750 14pages 2012

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



needs to make sure that the data used in any decision-making process originates from the correct source Dataauthentication prevents unauthorized parties from partici-pating in the networks and legitimate nodes should be ableto detect messages from authorized nodes and reject themIn [5] a robust user authentication scheme for WSNs wasproposed The scheme takes an advantage of the two-factorauthentication concept to provide a secure authenticationsystem offering balanced features in terms of security andperformance

The authenticity of data and commands is also a crit-ical requirement for the correct behavior of a WMSN[6] Digital images are becoming widely used in WMSNsas a kind of common multimedia information Thereforethe key establishment technique should guarantee that theimage communication and storage have a way for ver-ifying the authenticity creditability and integrity of thereceived image in WMSNs However resource constraintsin sensor networks (such as limited battery power andbandwidthcomputation capability) pose challenges for theimage authentication technique Conventional binary dataauthentication schemes could provide data integrity in astrict sense regardless of multimedia content However thoseschemes are not applicable toWMSNsbecause only simple biterrors during data transmission can lead to the authenticationfailing in spite of preserving multimedia content [7 8] Onthe other hand watermarking-based image content authenti-cation techniques are robust against bit errors packet lossesand compression distortionHowever watermark embeddingcreates extra source coding overheads and complicates trans-mission protocol design in WMSNs [9] which donrsquot adaptwell to WMSNs due to the constraints on limited energy andcomputing resources

In [10] an optimized content-aware authenticationscheme for JPEG 2000 images over lossy channels wasproposed An acyclic authentication graph was developed tooptimize the trade-off between the expected image distortionand the cryptohash tagging cost through the computationbased on packet loss probability and visual importance levelof the image packets The work reported in [11] proposeda JPEG 2000 compatible stream authentication scheme thatsignificantly reduced the computational complexity and hadonly a minimal authentication dependency overhead inWMSNs Moreover an authentication-aware wireless net-work resource allocation scheme was developed to reduceimage distortion and energy consumption during transmis-sion The scheme significantly improved the authenticatedimage quality even under strict communication energy con-sumption constraints in WMSN In [12] a rate-distortionoptimization authentication scheme for H264 video trans-missions was proposed A video packet transmission sched-uler was designed tominimize the visual distortion under thelimitation of total bit budget and authentication dependencyAnother related work regarding bite rrors robust imageor video authentication was given in [13ndash15] However allof these approaches are not able to be applied directly toWMSNs due to the energy constraint In this paper wepropose a perceptual hashing-based robust authenticationscheme for WMSNs Based on the distributed processing

strategy for perceptual image hashes the proposed schemecan provide compactness visual fragility perceptual robust-ness and security for image authentication in WMSNs

2 Perceptual Image Hashing Extraction

A perceptual image hashing function maps an image to ashort binary string as a digest based on an imagersquos appearanceto the human eye Perceptual image hashing is a class of one-way mappings from image presentations to a perceptual hashvalue in terms of their perceptual content Given an image 119868

and its perceptually similar copywithminor distortion 119868119889 theimage hashing function119867119896(sdot) depends on the secret key 119896 In[16] the desirable properties of perceptual hashing function119867119896(sdot) can be summarized as follows

(i) One-Way Function Ideally the hash generationshould be noninvertible

119868 997891997888rarr 119867119896 (119868) (1)

(ii) CompactnessThe size of the perceptual hashing valueshould be much smaller than that of the originalimage 119868

Size (119867119896 (119868)) ≪ Size (119868) (2)

(iii) Perceptual Robustness Perceptually identical imagesshould have similar perceptual hashing values

Pr 119867119896 (119868) asymp 119867119896 (119868119889) ge 1 minus 120576 0 le 120576 lt 1 (3)

(iv) Visual Fragility Perceptually distinct images shouldhave different perceptual hashing values

Pr 119867119896 (119868) =119867119896 (1198681015840) ge 1 minus 120591 0 le 120591 lt 1 (4)

(v) Security The perceptual hashing is intractable with-out the secret key

Pr 119867119896 (119868) =1198671198961015840 (119868) ge 1 minus 120575 0 le 120575 lt 1 (5)

All of the above parameters 120576 120591 and 120575 should be close to zeroBased on THE above properties perceptual image hash-

ing can be usually applied to image content identificationimage indexing content authentication and so forth Inparticular a perceptual hash function should have a propertythat is two images that look the same map to the samehash value even if the images have small bit-level differencesThis differentiates a perceptual hash from traditional cryp-tographic hashes such as SHA-l and MD5 In cryptographythe hash function is typically used for digital signature toauthenticate the message being sent so that the receivercan verify its source A key feature of conventional hashingalgorithms such as SHA-l andMD5 is that they are extremelysensitive to the input data that is changing even one bit ofthe input message will change the output dramatically How-ever image data often undergoes various content-preservingmanipulations such as lossy compression channel additive






119885119899+1 = 119906119885119899 (1 minus 119885119899) 119899 = 1 2 (6)

119878119899+1=(1 + 03 (119878119899minus1 minus 108) + 3791198782

119899+ 1001 times 119885

2

119899)mod 3

(7)





1= 0 lt


1015840(119894 119895) | 119894 = 1 2 1198711 119895 =


119910 = Ω (119889) =

[sum1198711

119894=1sum1198712

119895=1

10038161003816100381610038161003816119885 (119894 119895) minus 119885

1015840(119894 119895)

10038161003816100381610038161003816]

(1198711 times 1198712) (8)





14 and3 times 10



035

03

025

02

015

01

005

00 10

minus510minus10

10minus15

10minus20

10minus25

119910

119889

1198851119906


14

12

1

08

06

04

02

0

11987811198780

0 10minus5

10minus10

10minus15

10minus20

10minus25

119910

119889










119872119901119902 = int119868

119909119901119910119902119891 (119909 119910) 119889119909 119889119910 119901 119902 = 0 1 2 (9)


119872119901119902 = sum

(119909119910)isin119868

119909119901119910119902119891 (119909 119910) 119901 119902 = 0 1 2 (10)


119898119909 =11987210

11987200

119898119910 =11987201

11987200

(11)



119896

119909 119898119896


is calculated as

119898119896

119909=

sum119909sum119910119909 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119898119896

119910=

sum119909sum119910119910 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119896 = 1 2 119862

(12)




119872119903

10= sum

(119909119910)isinΩ

(119909 cos 120579 + 119910 sin 120579) 119891 (119909 119910)

= 11987210 cos 120579 + 11987201 sin 120579

119872119903

01= sum

(119909119910)isinΩ

(119910 cos 120579 minus 119909 sin 120579) 119891 (119909 119910)

= 11987201 cos 120579 minus 11987210 sin 120579

(13)



119866119903

119909=

119872119903

10

119872119903

00

= 119866119909 cos 120579 + 119866119910 sin 120579

119866119903

119910=

119872119903

01

119872119903

00

= 119866119910 cos 120579 minus 119866119909 sin 120579

(14)

Namely

[

[

119866119903

119909

119866119903

119910

]

]

= [cos 120579 sin 120579

minus sin 120579 cos 120579] [119866119909

119866119910] (15)




650

600

550

500

450

400

350

300

250

200100 200 300 400 500 600 700

119898119909

119898119910

(a) Rotate 30∘

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

119898119909

119898119910

(b) Scale 2 times

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910


0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910




(119909 119910) = 119877level minus 119891 (119909 119910) (16)




119896

119909=

sum119909sum119910119909 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

119896

119910=

sum119909sum119910119910 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

(17)




BS CH

1198751

1198752

1198753

1198754

119875119899

119878




119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)


119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)




119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

















119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation














119885119899+1 = 119906119885119899 (1 minus 119885119899) 119899 = 1 2 (6)

119878119899+1=(1 + 03 (119878119899minus1 minus 108) + 3791198782

119899+ 1001 times 119885

2

119899)mod 3

(7)





1= 0 lt


1015840(119894 119895) | 119894 = 1 2 1198711 119895 =


119910 = Ω (119889) =

[sum1198711

119894=1sum1198712

119895=1

10038161003816100381610038161003816119885 (119894 119895) minus 119885

1015840(119894 119895)

10038161003816100381610038161003816]

(1198711 times 1198712) (8)





14 and3 times 10



035

03

025

02

015

01

005

00 10

minus510minus10

10minus15

10minus20

10minus25

119910

119889

1198851119906


14

12

1

08

06

04

02

0

11987811198780

0 10minus5

10minus10

10minus15

10minus20

10minus25

119910

119889










119872119901119902 = int119868

119909119901119910119902119891 (119909 119910) 119889119909 119889119910 119901 119902 = 0 1 2 (9)


119872119901119902 = sum

(119909119910)isin119868

119909119901119910119902119891 (119909 119910) 119901 119902 = 0 1 2 (10)


119898119909 =11987210

11987200

119898119910 =11987201

11987200

(11)



119896

119909 119898119896


is calculated as

119898119896

119909=

sum119909sum119910119909 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119898119896

119910=

sum119909sum119910119910 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119896 = 1 2 119862

(12)




119872119903

10= sum

(119909119910)isinΩ

(119909 cos 120579 + 119910 sin 120579) 119891 (119909 119910)

= 11987210 cos 120579 + 11987201 sin 120579

119872119903

01= sum

(119909119910)isinΩ

(119910 cos 120579 minus 119909 sin 120579) 119891 (119909 119910)

= 11987201 cos 120579 minus 11987210 sin 120579

(13)



119866119903

119909=

119872119903

10

119872119903

00

= 119866119909 cos 120579 + 119866119910 sin 120579

119866119903

119910=

119872119903

01

119872119903

00

= 119866119910 cos 120579 minus 119866119909 sin 120579

(14)

Namely

[

[

119866119903

119909

119866119903

119910

]

]

= [cos 120579 sin 120579

minus sin 120579 cos 120579] [119866119909

119866119910] (15)




650

600

550

500

450

400

350

300

250

200100 200 300 400 500 600 700

119898119909

119898119910

(a) Rotate 30∘

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

119898119909

119898119910

(b) Scale 2 times

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910


0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910




(119909 119910) = 119877level minus 119891 (119909 119910) (16)




119896

119909=

sum119909sum119910119909 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

119896

119910=

sum119909sum119910119910 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

(17)




BS CH

1198751

1198752

1198753

1198754

119875119899

119878




119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)


119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)




119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

















119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation














119872119901119902 = int119868

119909119901119910119902119891 (119909 119910) 119889119909 119889119910 119901 119902 = 0 1 2 (9)


119872119901119902 = sum

(119909119910)isin119868

119909119901119910119902119891 (119909 119910) 119901 119902 = 0 1 2 (10)


119898119909 =11987210

11987200

119898119910 =11987201

11987200

(11)



119896

119909 119898119896


is calculated as

119898119896

119909=

sum119909sum119910119909 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119898119896

119910=

sum119909sum119910119910 sdot 119891119896(119909 119910)

sum119909sum119910119891119896 (119909 119910)

119896 = 1 2 119862

(12)




119872119903

10= sum

(119909119910)isinΩ

(119909 cos 120579 + 119910 sin 120579) 119891 (119909 119910)

= 11987210 cos 120579 + 11987201 sin 120579

119872119903

01= sum

(119909119910)isinΩ

(119910 cos 120579 minus 119909 sin 120579) 119891 (119909 119910)

= 11987201 cos 120579 minus 11987210 sin 120579

(13)



119866119903

119909=

119872119903

10

119872119903

00

= 119866119909 cos 120579 + 119866119910 sin 120579

119866119903

119910=

119872119903

01

119872119903

00

= 119866119910 cos 120579 minus 119866119909 sin 120579

(14)

Namely

[

[

119866119903

119909

119866119903

119910

]

]

= [cos 120579 sin 120579

minus sin 120579 cos 120579] [119866119909

119866119910] (15)




650

600

550

500

450

400

350

300

250

200100 200 300 400 500 600 700

119898119909

119898119910

(a) Rotate 30∘

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

119898119909

119898119910

(b) Scale 2 times

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910


0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910




(119909 119910) = 119877level minus 119891 (119909 119910) (16)




119896

119909=

sum119909sum119910119909 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

119896

119910=

sum119909sum119910119910 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

(17)




BS CH

1198751

1198752

1198753

1198754

119875119899

119878




119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)


119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)




119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

















119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










650

600

550

500

450

400

350

300

250

200100 200 300 400 500 600 700

119898119909

119898119910

(a) Rotate 30∘

0 200 400 600 800 10000

100

200

300

400

500

600

700

800

900

1000

119898119909

119898119910

(b) Scale 2 times

0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910


0 100 200 300 400 5000

50

100

150

200

250

300

350

400

450

500

119898119909

119898119910




(119909 119910) = 119877level minus 119891 (119909 119910) (16)




119896

119909=

sum119909sum119910119909 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

119896

119910=

sum119909sum119910119910 sdot exp (119891

119896(119909 119910) Δ 1)

sum119909sum119910exp (119891119896 (119909 119910) Δ 1)

(17)




BS CH

1198751

1198752

1198753

1198754

119875119899

119878




119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)


119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)




119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

















119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










BS CH

1198751

1198752

1198753

1198754

119875119899

119878




119863119896 (119866119896 119896)=max 10038161003816100381610038161003816119898119896

119909minus 119896

119909

1003816100381610038161003816100381610038161003816100381610038161003816119898119896

119910minus 119896

119910

10038161003816100381610038161003816 119896=1 2 119862

(18)


119863119896 = lfloor119863119896

Δ 2

rfloorΔ 2 (19)




119867 = (1198871111988712 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817(1198872111988722 sdot sdot sdot 1198871119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711989611198871198962 sdot sdot sdot 119887119896119895)2

10038171003817100381710038171003817sdot sdot sdot

10038171003817100381710038171003817(11988711986211198871198622 sdot sdot sdot 119887119862119895)2

(20)

















119863119867(1198671 1198672) =1

119871

119871

sum

119896=1

10038161003816100381610038161198671 (119896) minus 1198672 (119896)1003816100381610038161003816 (21)


Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Keys 1198961 1198962





Yes

No

Image isauthentic

Image isforged

DH(11986711198672) DH le 119879


0 01 02 03 04 05 060

10

20

30

40

50

60

70

80

minus01


Appe

aran

ce ti

me







0 5 10 15 20 25 300

005

01

015

02

025

03

035

04

Rotation angle

Nor

mal

ized

Ham

min

g di

stan

ce




119875119888 = 1 minus int

infin

119879

1

radic2120587120590

119890minus(119909minus120583)

221205902

119889119909 = 28106 times 10minus6 (22)




30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










30 40 50 60 70 80 900

005

01

015

02

025

03

035

04

045

05

Nor

mal

ized

Ham

min

g di

stan

ce

Quality factor


10 20 30 40 50 600

005

01

015

02

025

03

035

04

045

05

Noise variance

Nor

mal

ized

Ham

min

g di

stan

ce






Segment 1 (1198961= 39 119896





6 Conclusions


Acknowledgments



References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










References






















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article Perceptual Hashing-Based Robust Image...

Documents

Transcript of Research Article Perceptual Hashing-Based Robust Image...