Research Article An Improved Information Hiding …downloads.hindawi.com/journals/mpe/2015/197215.pdfResearch Article An Improved Information Hiding Method Based on Sparse Representation

Research ArticleAn Improved Information Hiding MethodBased on Sparse Representation

Minghai Yao123 Miao Qi1 Yugen Yi1 Yanjiao Shi1 and Jun Kong12

1School of Computer Science and Information Technology Northeast Normal University Changchun 130017 China2School of Mathematics and Statistics Northeast Normal University Changchun 130017 China3College of Information Science and Technology Bohai University Jinzhou 121013 China

Correspondence should be addressed to Miao Qi qim801nenueducn and Jun Kong kongj435nenueducn

Received 5 November 2014 Revised 9 December 2014 Accepted 9 December 2014

Academic Editor Hui Zhang

Copyright copy 2015 Minghai Yao 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 biometric authentication information hiding method based on the sparse representation is proposed for enhancing thesecurity of biometric information transmitted in the network In order to make good use of abundant information of the coverimage the sparse representation method is adopted to exploit the correlation between the cover and biometric images Thus thebiometric image is divided into two parts The first part is the reconstructed image and the other part is the residual image Thebiometric authentication image cannot be restored by any one part The residual image and sparse representation coefficients areembedded into the cover image Then for the sake of causing much less attention of attackers the visual attention mechanismis employed to select embedding location and embedding sequence of secret information Finally the reversible watermarkingalgorithm based on histogram is utilized for embedding the secret information For verifying the validity of the algorithm thePolyUmultispectral palmprint and the CASIA iris databases are used as biometric informationThe experimental results show thatthe proposed method exhibits good security invisibility and high capacity

1 Introduction

In recent years with the rapid development of the informa-tion technology Internet has become an indispensable part ofpeoplersquos lives In the meantime Internet fraud and networkattack have become a major problem to Internet usersNetwork service systemputs forward higher requirements forthe accuracy security and reliability of identity recognitionThe traditional identity recognition methods such as smartcard ID and password have been unable to meet the needof people Biometric characteristics of the human body suchas palmprint iris and face have the properties of uniquenessand invariability which have become an important founda-tion for identity recognition

The biometrics is a new identification technique Anindividual is identified by their distinct physiological orbehavioral characteristics The identification method basedon biometric technology is better than traditional methodsHowever biometric data itself has no confidentiality and

security Therefore the security problem of biometric datahas already become an urgent and important problem Infor-mation hiding is an effective solution to protect security andintegrity of biometric data Many researchers have proposedvarious methods for protecting biometric data

Bedi et al presented the multimodal biometric authenti-cation method using PSO based watermarking [1] The keyidea is that the multimodal biometric image was used as thewatermark image PSO was used to select best DCT coeffi-cient in the face image Vatsa et al presented a three-levelRDWT biometric watermarking algorithm for embeddingthe voice biometric MFC coefficient into a color face image[2] The watermarking algorithm used adaptive user-specificparameters for improving performance A novel method ofempirical mode decomposition and gene expression pro-gramming together was used to embed biometric informa-tion in the literature [3] The singular value decompositionand lifting based discrete wavelet transform were employedin the watermarking algorithm Li et al proposed Tamper

Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2015 Article ID 197215 10 pageshttpdxdoiorg1011552015197215

2 Mathematical Problems in Engineering

Dictionary

Cover image

Biometric image

Sparserepresentation

Reconstructedimage

Residualimage

Reconstructedcoefficient

Hiding

Secret key

Image extraction

Secret key

Residualimage


Cover imageDictionary

Reconstructedimage

Biometric image

Vision attention

Stego-image

Figure 1 The flowchart of biometric information hiding method

detection and self-recovery of biometric images using salientregion-based authentication watermarking scheme [4] Cheet al took into account the content relevance of the coverimage and the watermarked image they proposed content-based image hiding method [5] The biometric informationhiding methods based on correlation analysis were proposedin the literature [6] For the video data the dual videowatermark authentication method has been proposed by Shiet al [7]

Through analyzing existing biometric information hidingmethods we found almost all of hidingmethods adopt digitalwatermarking and hide one or more biometric images ortheir features into another biometric image directly based ontransform domain for verification These methods are robustagainst some types of attacks but the hiding capacity is lowAt the same time the existing methods rarely consider thecontent correlation between the biometric image and thecover imageThe cover image is only used as a hidden carrierand their rich content cannot be fully utilized Therefore anovel biometric authentication image hiding method basedon the sparse representation is proposed in this paper whichconsiders the content correlation between the biometricimage and the cover image adequately

2 The Proposed Method

The sparse representation and visual attention model areused in the biometric hiding method which uses the sparserepresentation theory to analyze the content correlationbetween the cover image and the biometric image Firstthe dictionary is built by the cover images for calculatingthe sparse representation coefficients of a biometric imageThe biometric image is reconstructed by the dictionary and

sparse representation coefficientsThe difference between theoriginal biometric image and the reconstructed biometricimage is used as one part of the secret information Thesparse representation coefficients are another part of thesecret information The two parts of secret information areembedded into the cover image In order to facilitate theembedding of secret information the secret information isconverted into a binary sequenceThrough statistical analysisof a large number of experimental data we found that theabsolute value of each pixel value in the residual image isless than 31 Therefore each pixel value of residual imagecan be represented by six binary bits where the first onerepresents the sign bit and the other five are the pixel valuesThe amount of embedded information is greatly reducedcompared with the original image Besides the embeddingregion of secret information is selected by the saliency imageof cover image At last the reversible watermarking algorithmbased on histogram is used to embed secret informationwhere the embedded order is determined by the saliencyimage of the cover image

When we extract the secret information the binarysequence in stego-image is firstly extracted by the secretkey Then the binary sequence is converted to residualimage and sparse representation coefficients The dictionaryis built by recovered stego-image Finally the biometric imageis restored by the residual image and the reconstructedbiometric image

The flowchart of hiding method is described in Figure 1The method is divided into three stages of sparse representa-tion vision attention and image hiding

21 Sparse Representation With the rapid development ofcomputer technology abundant signal processing methods

Mathematical Problems in Engineering 3

have been proposed As a branch of signal processing sparserepresentation has been wildly applied in image denoisingimage restoration feature extraction image compressionpattern recognition machine learning compressed sensingand other fields Nevertheless the sparse representationtheory is rarely used in the information hiding field Jost et alapplied the sparse representation theory to the informationsecurity field the secret information was embedded into thedecomposition path of the sparse decomposition image Thereceiver extracts the secret information by decompositionpath of the cover image [8] Cencelli and his colleaguesembedded information by modifying the sparse represen-tation coefficients of image [9ndash11] These methods only usea sparse representation coefficient and decomposition pathto achieve information hiding They do not analyze thecorrelation between the secret information and the coverimage by sparse representationmethodTherefore this paperuses sparse representation method to analyze the correlationbetween secret information and the cover image for reducingthe number of secret information

The element of decomposition signal is called the atom inthe sparse representation theory Suppose 119910 isin 119877119899 is an imageblock representation vector 119863 isin 119877119899times119871 represents redundantdictionary 119871 represents the number of atoms and sparserepresentation process can be described as

= arg min120572

1003817100381710038171003817119910 minus 11986312057210038171003817100381710038172

2st 120572

0le 119878 (1)

where 1003817100381710038171003817lowast10038171003817100381710038172 represents the 1198972-norm 1003817100381710038171003817lowast

10038171003817100381710038170 represents the 1198970-norm 119878 represents sparse degree 119863 is the redundant dictio-nary119863 = [119889

1 1198892 119889

119871] 120572 represents sparse representation

coefficient and 100381710038171003817100381712057210038171003817100381710038170 represents the number of non-zero

coefficients in 120572 Figure 2 is schematic diagram of the sparserepresentation theory

Two of the most critical problems need to be solvedusing the sparse representation for information hiding Oneis how to build an effective dictionary 119863 and the other ishow to obtain sparse representation coefficient Taking intoaccount the characteristics of the cover image and biometricimage we use the integer value dictionary In order to reducethe difficulty and the complexity of building dictionary datasamplingmethod is used to build the redundant dictionary bythe cover image Palmprint images and iris images are dividedinto blocks for improving the computational efficiency of thealgorithm in this paper Palmprint image is divided into 16blocks with the size of 32 times 32 iris image is divided into 8blocks with the size of 32 times 32 The dictionary is built bydata sampling methods with size of 1024 times 6561 Orthogonalmatching pursuit algorithm (OMP) is used to calculate thesparse representation coefficient [12] which is an improvedalgorithm based on matching pursuit algorithm [13] Theoriginal image reconstructed image and residual image areshown in Figures 3 and 4

22 Vision Attention Visual attention mechanism is anemerging research field which contains neurobiology psy-chology computer vision pattern recognition artificial intel-ligence and many other disciplines It is one kind of mecha-nism of human visual systems in selecting regions of interest

y D

a

times=

Figure 2 The schematic diagram of the sparse representationtheory

from complex scenes Recently it has become the focus ofresearch in computer vision due to its applications for objectdetection [14ndash16] and digital image processing field [17ndash20]Visual attentionmodel is divided into two categories by visualinformation processing method in computer vision systemThe first one is a bottom-up visual attention model methodwhich is directed by the data But it is not dependent on thespecific task The second one is a top-down visual attentionmodel method which is affected by subjective consciousnessIt depends on the specific task In this paper we use themethod of literature [21]The saliency of object is detected bythe unified approach which integrates bottom-up for lower-level features and top-down for higher-level priors

After multiscale feature extraction we decompose animage into small regions by image segmentation Then themean of the feature vectors in a segment is treated as thefeature The matrix representation of the image is formed bystacking them By this means even when the object size islarge the number of segments in a salient object is still smalldue to the fact that salient objects usually have spatial andappearance-wise coherence At the same time in order toensure that the matrix representing the background has a lowrank we train a linear feature transformation using labeleddataThis method can achieve good performance on saliencydetection even without higher-level knowledge Figures 5and 6 show the illustration of the method

In order to ensure the consistency of the saliency imagefrom the cover image and the stego-image the saliencyimage is computed by the reference subsampling image Thereference subsampling image is computed by (3) and (4)

23 Information Hiding One target of information hidingis to hide secret information into another nonsecret coverimage for avoiding the attackerrsquos attention The informationhiding technique requires that the secret information cannotbe found in digital media At the same time if the attackerdoes not get the secret key anyone cannot extract the secretinformation from the digital media


Original image Sparse representation results Residual image

Figure 3 Iris image reconstruction result


Figure 4 Palmprint image reconstruction result

Cover image Image saliency

Figure 5 Lena image and its saliency image

The histogram is the most basic statistical characteristicsof the image A digital image has 119871 gray-level in the range[0 119866] and the discrete function of the histogram is definedas

ℎ (119896) = 119899119896 (2)

where 119896 represents the gray value of the image 119899119896represents

the number of the pixel that the gray value is 119896 and the valueof 119866 is 255 in the gray image

Histogram analysis is an important tool for digital imageprocessing Thus the reversible watermarking algorithmbased onhistogram is used in the process of embedding secret



Figure 6 Airplane image and its saliency image

information The research focus of the traditional hidingalgorithm based on histogram is how to determine the peak-value point and zero-value point which results in that thewatermark embedding capacity is too small and randomThe application scope of the algorithm is limited because ofthese problems A large number of statistical results showthat the embedding capacity can be increased by segmentingthe image into blocks Since the pixel value is relativelyconcentrated in the image block more embedded space canbe got from the image [22]

The watermark embedding method of literature [23] isused in this paper meanwhile the division block methodof cover image and embedding sequence of watermark areimproved The cover image is divided into small pieces bysaliency image each of which is a small cover image Allimage blocks are sorted by significance The secret informa-tion is segmented according to the embedding capacity ofeach small cover image and embedded in the correspondingcover image block until all the secret information is embed-ded

The embedding method of secret information can bedescribed as follows

Step 1 According to the significance the block of cover image119868 is selected and the values of sampling coefficients 119906 and Vare 2 respectively All the subsampled images 119868

1 1198682 1198683 1198684are

generated by (3)

119868119898(119894 119895) = 119868 (119894 sdot V + floor(119898 minus 1

119906) 119895 sdot 119906 + (119898 minus 1) mod 119906)

(3)

Step 2 119868Ref represents the reference subsampled image 119868Desrepresents the target subsampled image 119868RefminusDes representsthe difference of 119868Ref and 119868Des All 119868RefminusDes are calculated by (5)

119868Ref = (Round(119906

2minus 1)) times V + Round( V

2) (4)

119868RefminusDes (119894 119895) = 119868Ref (119894 119895) minus 119868Des (119894 119895) (5)

where 0 le 119894 le 119872(V minus 1) 0 le 119895 le 119873(119906 minus 1)

Step 3 The embedding position of secret information isdetermined bymoving the histogrambased on the embeddedlevel 119871 In order to adaptively embed secret information thehistogram is moved around according to embedding level 119871When the histogram is modified the secret information isembedded in the range of [minus119871 119871] The moving method of thehistogram is shown in the following

119867119878= 119867 + 119871 + 1 119867 ge 119871 + 1

119867 minus 119871 minus 1 119867 le minus119871 minus 1(6)

where119867 represents the gray values and 119871 represents embed-ding level

The pixel value of the reference subsampling imagecannot be changed in order to ensure reversibility of themethod Therefore we can only modify the pixel value ofthe target subsampled imageThemodificationmethod of thepixel value is shown in the following

1198681015840

Des (119894 119895) = 119868Des (119894 119895) minus (119871 + 1) 119867 ge 119871 + 1

119868Des (119894 119895) + (119871 + 1) 119867 le minus119871 minus 1(7)

Step 4 The secret information is embedded by moving thehistogram We scan the pixel value of 119868RefminusDes When thesize of the pixel value is 119871 or minus119871 the secret information isembeddedThe scanning process is repeated until all the pixelvalues are not equal to 119871 or minus119871 In this case the embeddedlevel 119871 is minus1 and repeat the above process until 119871 lt 0 Theembedding method is shown in (8) The moving process ofthe histogram is shown in Figure 7

119871 gt 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) minus (119871 + 1) 1198751015840

= 119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + (119871 + 1) 1198751015840

= minus119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + 119871 1198751015840

= minus119871 119908 (119899) = 0

1198681015840

Des (119894 119895) minus 119871 1198751015840

= 119871 119908 (119899) = 0


minus6 minus5 minus4 minus3 minus2 minus1 0 1 2 3 4 5 6 minus6 minus5 minus4 minus3 minus2 minus1 0 1 2 3 4 5 6

0

1

0

1

0

1

0

1

Figure 7 The moving process of histogram

Cover image Stego-image

Figure 8 The cover images and the stego-images contain iris information

119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)

The original image and the stego-image are shown inFigures 8 and 9 The extraction of secret information is aninverse process of the embedding process First the referencesubsampled image is extracted from stego-image for com-puting the saliency image Then the stego-image is dividedinto blocksThe image blocks are sorted by the significance ofimage blockThe reference subsampled image of image blockis extracted by (3) and (4)The secret information is extracted

by the secret key and the reference subsampled image Finallythe biometric authentication image is reconstructed by resid-ual image dictionary and sparse representation coefficients

3 Experiment and Analysis

The proposed method is verified by the biometric authenti-cation data in this paper and the performance of the methodis tested from security invisibility and embedding capacity

31 Experimental Data The PolyU multispectral palmprintdatabase and the CASIA iris database of Chinese Academy ofSciences are used as biometric information [24 25]We select100 images from each database The size of palmprint image



Figure 9 The cover images and the stego-images contain palmprint information

Figure 10 The iris images

is 128 times 128 the size of iris image is 128 times 64 Some images inthe database are shown in Figures 10 and 11The cover imagescontain rich texture information and are unrelated with thebiometric image The cover images are shown in Figure 12

32 Performance Analysis Firstly from the security point ofview the biometric information is hidden into the unnoticedcover image which reduces the attackerrsquos attention Secondlythe main part of the biometric image is reconstructed by thesparse representation method and the other part is embed-ded into the cover image Even if the embedded informationis intercepted the attacker cannot restore the entire biometricimage Finally the visual attention mechanism is used toselect embedding location and embedding sequence of secretinformation The visual attention mechanism increases theconfidentiality of embedded information

Peak Signal to Noise Ratio (PSNR) is an effective way toevaluate invisibility of information hiding The PSNR of animage is calculated by the following

PSNR = 10 times log10(

255 times 255 times119872 times119873

sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)

where 119862 represents the cover image and 119878 represents thestego-image When PSNR is higher than 30 dB we believethat the stego-image holds good invisibility The proposedmethod and the method of literature [26] are compared andthe comparison results are shown in Table 1

From Table 1 we can see that the invisibility of ourmethod is better than the other Because the correlationanalysis method based on sparse representation is used in theinformation hiding the PSNR value of our method is higher


Table 1 PSNR comparison results

Cover image Cover image (pixel) Database PSNR (dB)Literature [26] The proposed method

Lena 512 times 512 CASIA 3652 4197Lena 512 times 512 PolyU 3426 3762Airplane 512 times 512 CASIA 385 4490Airplane 512 times 512 PolyU 3442 3942Barbara 512 times 512 CASIA 3612 4050Barbara 512 times 512 PolyU 3305 3605

Figure 11 The palmprint images

Figure 12 The cover images

than the other method At the same time the cover image ischanged very little in the embedding process due to the factthat the pixel value of the residual image is very small Theembedding information has a minor effect on cover image

The embedding capacity is a main evaluation criterionfor information hiding algorithmThe embedding capacity iscalculated by the following

bpp =119873119904

119872119888times 119873119888

(10)

where119873119904represents the number of binary bits of secret infor-

mation and119872119888times119873119888represents the size of the cover imageThe

proposed method and the method of literature [27ndash29] are

compared to embedding capacity and the comparison resultsare shown in Table 2

Table 2 shows the embedding capacity comparison resultsof different methodsThe embedding capacity of our methodis higher than other methods Relatively speaking Vatsa et alget the worst results

4 Conclusions

A novel biometric image hiding method based on the sparserepresentation is proposed in this paper The transmissionsecurity of biometric information is improved in the networkThe biometric image is divided into the reconstructed imagewith high energy and the residual image with low energy


Table 2 Embedding capacity comparison results

Algorithm Cover image (pixel) Secret information (bit) bppVatsa et al [27] 1024 times 768 1000 00013Vatsa et al [28] 512 times 512 16384 00625Shih and Wu [29] 512 times 512 times 3 86016 01094Our method 512 times 512 142084 05420

Residual image is embedded into the cover image In order toreduce the attackerrsquos attention to stego-image visual attentionmechanism is used in the secret information embedding pro-cess At the same time the hiding strategy is modified in theinformation hiding algorithm where the secret informationembedding process is guided by the saliency image Since thereconstruction method has a certain complexity the secretkey has high secrecy and is difficult to decipher Because thebiometric image is divided into two parts the attacker cannotrestore biometric authentication images with any part Alarge number of experimental results show that the proposedmethod can protect the biometric authentication informationeffectively Currently our approach can only process the gray-scale image How to hide biometric information into colorimage is an important problem in future research

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the Science and TechnologyResearch Project of Liaoning Province Education Depart-ment (no L2014450) Social Science Planning FoundationProject of Liaoning Province (no L13BXW006) Fund ofJilin Provincial Science amp Technology Department (nos20130206042GX and 20130522115JH) and National NaturalScience Foundation of China (no 61403078)

References

[1] P Bedi R Bansal and P Sehgal ldquoMulitimodal biometricauthentication using PSO based watermarkingrdquo Procedia Tech-nology vol 4 pp 612ndash618 2012

[2] M Vatsa R Singh and A Noore ldquoFeature based RDWTwatermarking for multimodal biometric systemrdquo Image andVision Computing vol 27 no 3 pp 293ndash304 2009

[3] A K Shaw S Majumder S Sarkar and S K Sarkar ldquoA novelEMD based watermarking of fingerprint biometric using GEPrdquoProcedia Technology vol 10 pp 172ndash183 2013

[4] C Li Y Wang B Ma and Z Zhang ldquoTamper detection andself-recovery of biometric images using salient region-basedauthenticationwatermarking schemerdquoComputer Standards andInterfaces vol 34 no 4 pp 367ndash379 2012

[5] X Che J Kong J Dai Z Gao andMQi ldquoContent-based imagehiding method for secure network biometric verificationrdquoInternational Journal of Computational Intelligence Systems vol4 no 4 pp 596ndash605 2011

[6] M Qi Y Lu N Du Y Zhang C Wang and J Kong ldquoAnovel image hiding approach based on correlation analysisfor secure multimodal biometricsrdquo Journal of Network andComputer Applications vol 33 no 3 pp 247ndash257 2010

[7] Y Shi M Qi Y Yi M Zhang and J Kong ldquoObject based dualwatermarking for video authenticationrdquo Optik vol 124 no 19pp 3827ndash3834 2013

[8] P Jost P Vandergheynst and P Frossard ldquoRedundant imagerepresentations in security applicationsrdquo in Proceedings of theInternational Conference on Image Processing (ICIP rsquo04) pp2151ndash2154 October 2004

[9] G Cancelli M Barni and G Menegaz ldquoMPSteg hiding amessage in thematching pursuit domainrdquo in Security Steganog-raphy andWatermarking ofMultimedia Contents VIII vol 6072of Proceedings of SPIE pp 1ndash9 2006

[10] G Cancelli andM Barni ldquoMPSteg-color a new steganographictechnique for color imagesrdquo in Information Hiding 9th Inter-national Workshop (IH rsquo07) Saint Malo France June 11ndash132007 vol 4567 of Lecture Notes in Computer Science pp 1ndash15Springer Berlin Germany 2007

[11] G Cancelli and M Barni ldquoMPSteg-color data hiding throughredundant basis decompositionrdquo IEEETransactions on Informa-tion Forensics and Security vol 4 no 3 pp 346ndash358 2009

[12] Y C Pati R Rezaiifar and P S Krishnaprasad ldquoOrthogonalmatching pursuit recursive function approximation with appli-cations to wavelet decompositionrdquo in Proceedings of the 27thAnnul Asilomar Conference on Signals Systems and Computerspp 40ndash44 Pacific Grove Calif USA 1993

[13] S G Mallat and Z Zhang ldquoMatching pursuits with time-frequency dictionariesrdquo IEEE Transactions on Signal Processingvol 41 no 12 pp 3397ndash3415 1993

[14] MWGuo Y Z Zhao C B Zhang and ZH Chen ldquoFast objectdetection based on selective visual attentionrdquo Neurocomputingvol 144 no 20 pp 184ndash197 2014

[15] H P Yu Y X Chang P Lu Z Y Xu C Y Fu and Y FWang ldquoEfficient object detection based on selective attentionrdquoComputers and Electrical Engineering vol 40 no 3 pp 907ndash9192014

[16] Q Liu Q Z Zhang W B Chen and Z C Huang ldquoPedestriandetection based on modeling computation of visual attentionrdquoJournal of Beijing Information Science amp Technology Universityvol 29 no 2 pp 59ndash65 2014

[17] C Guo and L Zhang ldquoA novel multiresolution spatiotemporalsaliency detectionmodel and its applications in image and videocompressionrdquo IEEE Transactions on Image Processing vol 19no 1 pp 185ndash198 2010

[18] C Yang L H Zhang H C Lu X Ruan and M-H YangldquoSaliency detection via graph-based manifold rankingrdquo inProceedings of the 26th IEEEConference onComputer Vision andPattern Recognition (CVPR rsquo13) pp 3166ndash3173 Portland OreUSA June 2013


[19] B Yang and S T Li ldquoVisual attention guided image fusion withsparse representationrdquo Optik vol 125 no 17 pp 4881ndash48882014

[20] H-Y Yang Y-W Li W-Y Li X-Y Wang and F-Y YangldquoContent-based image retrieval using local visual attentionfeaturerdquo Journal of Visual Communication and Image Represen-tation vol 25 no 6 pp 1308ndash1323 2014

[21] X H Shen Y Wu and Evanston ldquoA unified approach tosalient object detection via low rank matrix recoveryrdquo inProceedings of the IEEE Conference on Computer Vision andPattern Recognition (CVPR rsquo12) pp 853ndash860 Providence RIUSA June 2012

[22] K-S Kim M-J Lee H-Y Lee and H-K Lee ldquoReversibledata hiding exploiting spatial correlation between sub-sampledimagesrdquo Pattern Recognition vol 42 no 11 pp 3083ndash30962009

[23] J H Hwang J W Kim and J U Choi ldquoA reversible water-marking based on histogram shiftingrdquo inDigitalWatermarking5th International Workshop (IWDW rsquo06) Jeju Island Republicof Korea November 8ndash10 2006 vol 4283 of Lecture Notesin Computer Science pp 348ndash361 Springer Berlin Germany2006

[24] PolyUmultispectral palmprint Database httpwwwcomppo-lyueduhksimbiometricsMultispectralPalmprintMSPhtm

[25] CASIA-IrisV1[DBOL] 2004 httpwwwcbsriaaccnIrisDa-tabasehtm

[26] M Qi J Y Dai J Z Wang X X Yu and M Zhang ldquoContent-based reversible steganographic method for multimodal bio-metricsrdquo Computer Science vol 39 no 11 pp 70ndash74 2012

[27] M Vatsa R Singh P Mitra et al ldquoComparing robustness ofwatermarking algorithms on biometrics datardquo in Proceedings ofthe Workshop on Biometric Challenges from Theory to Practice(ICPR Workshop rsquo04) pp 5ndash8 August 2004

[28] M Vatsa R Singh A Noore M M Houck and K MorrisldquoRobust biometric image watermarking for fingerprint and facetemplate protectionrdquo IEICE Electronics Express vol 3 no 2 pp23ndash28 2006

[29] F Y Shih and S Y T Wu ldquoCombinational image watermakingin the spatial and frequency domainsrdquo Pattern Recognition vol36 no 4 pp 969ndash975 2002

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


Dictionary

Cover image

Biometric image

Sparserepresentation

Reconstructedimage

Residualimage


Hiding

Secret key

Image extraction

Secret key

Residualimage


Cover imageDictionary

Reconstructedimage

Biometric image

Vision attention

Stego-image

Figure 1 The flowchart of biometric information hiding method

detection and self-recovery of biometric images using salientregion-based authentication watermarking scheme [4] Cheet al took into account the content relevance of the coverimage and the watermarked image they proposed content-based image hiding method [5] The biometric informationhiding methods based on correlation analysis were proposedin the literature [6] For the video data the dual videowatermark authentication method has been proposed by Shiet al [7]

Through analyzing existing biometric information hidingmethods we found almost all of hidingmethods adopt digitalwatermarking and hide one or more biometric images ortheir features into another biometric image directly based ontransform domain for verification These methods are robustagainst some types of attacks but the hiding capacity is lowAt the same time the existing methods rarely consider thecontent correlation between the biometric image and thecover imageThe cover image is only used as a hidden carrierand their rich content cannot be fully utilized Therefore anovel biometric authentication image hiding method basedon the sparse representation is proposed in this paper whichconsiders the content correlation between the biometricimage and the cover image adequately

2 The Proposed Method

The sparse representation and visual attention model areused in the biometric hiding method which uses the sparserepresentation theory to analyze the content correlationbetween the cover image and the biometric image Firstthe dictionary is built by the cover images for calculatingthe sparse representation coefficients of a biometric imageThe biometric image is reconstructed by the dictionary and

sparse representation coefficientsThe difference between theoriginal biometric image and the reconstructed biometricimage is used as one part of the secret information Thesparse representation coefficients are another part of thesecret information The two parts of secret information areembedded into the cover image In order to facilitate theembedding of secret information the secret information isconverted into a binary sequenceThrough statistical analysisof a large number of experimental data we found that theabsolute value of each pixel value in the residual image isless than 31 Therefore each pixel value of residual imagecan be represented by six binary bits where the first onerepresents the sign bit and the other five are the pixel valuesThe amount of embedded information is greatly reducedcompared with the original image Besides the embeddingregion of secret information is selected by the saliency imageof cover image At last the reversible watermarking algorithmbased on histogram is used to embed secret informationwhere the embedded order is determined by the saliencyimage of the cover image

When we extract the secret information the binarysequence in stego-image is firstly extracted by the secretkey Then the binary sequence is converted to residualimage and sparse representation coefficients The dictionaryis built by recovered stego-image Finally the biometric imageis restored by the residual image and the reconstructedbiometric image

The flowchart of hiding method is described in Figure 1The method is divided into three stages of sparse representa-tion vision attention and image hiding

21 Sparse Representation With the rapid development ofcomputer technology abundant signal processing methods




= arg min120572

1003817100381710038171003817119910 minus 11986312057210038171003817100381710038172

2st 120572

0le 119878 (1)



1 1198892 119889






y D

a

times=














ℎ (119896) = 119899119896 (2)











1 1198682 1198683 1198684are

generated by (3)


119906) 119895 sdot 119906 + (119898 minus 1) mod 119906)

(3)


119868Ref = (Round(119906


2) (4)




119867119878= 119867 + 119871 + 1 119867 ge 119871 + 1




1198681015840

Des (119894 119895) = 119868Des (119894 119895) minus (119871 + 1) 119867 ge 119871 + 1



119871 gt 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) minus (119871 + 1) 1198751015840

= 119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + (119871 + 1) 1198751015840

= minus119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + 119871 1198751015840

= minus119871 119908 (119899) = 0

1198681015840

Des (119894 119895) minus 119871 1198751015840

= 119871 119908 (119899) = 0



0

1

0

1

0

1

0

1




119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra












= arg min120572

1003817100381710038171003817119910 minus 11986312057210038171003817100381710038172

2st 120572

0le 119878 (1)



1 1198892 119889






y D

a

times=














ℎ (119896) = 119899119896 (2)











1 1198682 1198683 1198684are

generated by (3)


119906) 119895 sdot 119906 + (119898 minus 1) mod 119906)

(3)


119868Ref = (Round(119906


2) (4)




119867119878= 119867 + 119871 + 1 119867 ge 119871 + 1




1198681015840

Des (119894 119895) = 119868Des (119894 119895) minus (119871 + 1) 119867 ge 119871 + 1



119871 gt 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) minus (119871 + 1) 1198751015840

= 119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + (119871 + 1) 1198751015840

= minus119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + 119871 1198751015840

= minus119871 119908 (119899) = 0

1198681015840

Des (119894 119895) minus 119871 1198751015840

= 119871 119908 (119899) = 0



0

1

0

1

0

1

0

1




119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra

















ℎ (119896) = 119899119896 (2)











1 1198682 1198683 1198684are

generated by (3)


119906) 119895 sdot 119906 + (119898 minus 1) mod 119906)

(3)


119868Ref = (Round(119906


2) (4)




119867119878= 119867 + 119871 + 1 119867 ge 119871 + 1




1198681015840

Des (119894 119895) = 119868Des (119894 119895) minus (119871 + 1) 119867 ge 119871 + 1



119871 gt 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) minus (119871 + 1) 1198751015840

= 119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + (119871 + 1) 1198751015840

= minus119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + 119871 1198751015840

= minus119871 119908 (119899) = 0

1198681015840

Des (119894 119895) minus 119871 1198751015840

= 119871 119908 (119899) = 0



0

1

0

1

0

1

0

1




119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















1 1198682 1198683 1198684are

generated by (3)


119906) 119895 sdot 119906 + (119898 minus 1) mod 119906)

(3)


119868Ref = (Round(119906


2) (4)




119867119878= 119867 + 119871 + 1 119867 ge 119871 + 1




1198681015840

Des (119894 119895) = 119868Des (119894 119895) minus (119871 + 1) 119867 ge 119871 + 1



119871 gt 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) minus (119871 + 1) 1198751015840

= 119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + (119871 + 1) 1198751015840

= minus119871 119908 (119899) = 1

1198681015840

Des (119894 119895) + 119871 1198751015840

= minus119871 119908 (119899) = 0

1198681015840

Des (119894 119895) minus 119871 1198751015840

= 119871 119908 (119899) = 0



0

1

0

1

0

1

0

1




119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











0

1

0

1

0

1

0

1




119871 = 0

11986810158401015840

Des (119894 119895) =

1198681015840

Des (119894 119895) 1198751015840

= 0 119908 (119899) = 1

1198681015840

Des (119894 119895) minus 1 1198751015840

= 0 119908 (119899) = 0

(8)















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra


















sum119872

119894=1sum119873

119895=1[119862(119894 119895) minus 119878(119894 119895)]

2) (9)











bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra

















bpp =119873119904

119872119888times 119873119888

(10)






4 Conclusions








Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra















Acknowledgments


References






































Volume 2014




Journal of











Journal of


Function Spaces






Algebra




























Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Volume 2014




Journal of











Journal of


Function Spaces






Algebra









Documents

Research Article An Improved Information Hiding …downloads.hindawi.com/journals/mpe/2015/197215.pdfResearch Article An Improved Information Hiding Method Based on Sparse Representation