International
Journal
of
Electronics
and
Communications
(AEÜ)
jo u rn al h om e p a g e :w w w . e l s e v i e r . d e / ae u e
Reversible
data
hiding
based
on
multilevel
histogram
modification
and
sequential
recovery
Zhenfei
Zhao
a,b,
Hao
Luo
c,∗,
Zhe-Ming
Lu
c,
Jeng-Shyang
Pan
d aSchoolofInformationScienceandTechnology,SunYat-senUniversity,Guangzhou,ChinabHeilongjiangInstituteofScienceandTechnology,Harbin,China
cSchoolofAeronauticsandAstronautics,ZhejiangUniversity,No.38ZheDaRoad,Hangzhou310027,China dDepartmentofElectronicEngineering,NationalKaohsiungUniversityofAppliedSciences,Kaohsiung,Taiwan
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received14January2010 Accepted21January2011
Keywords:
Reversibledatahiding
Multilevelhistogrammodification Sequentialrecovery
a
b
s
t
r
a
c
t
Thispaperproposesareversibledatahidingmethodfornaturalimages.Duetothesimilarityofneighbor pixels’values,mostdifferencesbetweenpairsofadjacentpixelsareequalorclosetozero.Inthiswork, ahistogramisconstructedbasedonthesedifferencestatistics.Inthedataembeddingstage,amultilevel histogrammodificationmechanismisemployed.Asmorepeakpointsareusedforsecretbitsmodulation, thehidingcapacityisenhancedcomparedwiththoseconventionalmethodsbasedononeortwolevel histogrammodification.Moreover,asthedifferencesconcentricityaroundzeroisimproved,the distor-tionsonthehostimageintroducedbysecretcontentembeddingismitigated.Inthedataextractionand imagerecoverystage,theembeddinglevelinsteadofthepeakpointsandzeropointsisused. Accord-inglytheaffiliatedinformationismuchsmallerthaninthosemethodsofthekind.Asequentialrecovery strategyisexploitedforeachpixelisreconstructedwiththeaidofitspreviouslyrecoveredneighbor. Experimentalresultsandcomparisonswithothermethodsdemonstrateourmethod’seffectivenessand superiorperformance.
© 2011 Elsevier GmbH. All rights reserved.
1. Introduction
Datahiding,alsocalledinformationhiding,playsanimportant roleininformationsecurity.Itaimsatembeddingimperceptible confidentialinformationincovermediasuchasstillimages,videos, audios,3D meshes, etc. It consists of severalbranches suchas steganography,watermarking,visualcryptography,etc.Thedata hidingschemeproposedinthisworkcanbeclassifiedintothe cat-egoryofsteganography.Steganographyisusuallyusedforcovert communications.Thusthehighembeddingcapacity isthemain concernin this kindof technique. In contrast,watermarking is usually usedfor copyright protection and announcement. Thus researchersaimatimprovingtherobustnessofwatermark con-tentagainstintentionalorunintentionalattacks.Therefore,most availabledatahidingmethodscanprovideahighercapacitythan thatprovidedbywatermarkingschemes.Thisadvantagebroadens theapplicationscenariosofdatahiding.
Nowadays,variousdatahidingtechniqueshavebeenreportedin literatures[1,2,29–31].Asaburgeoningbranch,reversibledata hid-inghasdrawnmuchattentionamongresearchers.Itskeyproperty isnotonlythesecretdatabutalsothehostimagecanbeaccurately
∗ Correspondingauthor.Tel.:+8615858259064;fax:+8657186971612. E-mailaddress:luohao@zju.edu.cn(H.Luo).
recoveredindecoder.Therefore,itcanbeusedinthoseapplications wherethehostimages(e.g.,militarymaps,remotesensingimages, medicalimages[11],digitalizedartpictures,etc.)mustbeexactly reconstructed.Incontrast,theconventionalirreversibledatahiding methodsarenotappropriateanylonger.
Availablereversibledatahidingtechniquescanbedividedinto spatialdomain,transformdomainandcompresseddomain meth-ods.Inthespatialdomainbasedmethods[3–20],thesecretdatais usuallyembeddedbypixels’valuesmodification.Inthetransform domainmethods,somereversibility-guaranteedtransforms(e.g., integerdiscretecosinetransform[21,22],integerwavelet trans-form[23])are exploitedand thedata embeddingisreduced to coefficientsmodulation.Inthecompresseddomainmethods,some popularusedimagecompressiontechniques(e.g.,vector quantiza-tion[24–26],blocktruncationcoding[27],MPEGcoding[28])are involved.
Most spatial domain reversible data hiding are developed basedontwoprinciples,i.e.,differenceexpansion(DE)[3–12]and histogram modification [13–20]. In general,the former kindof methodscanprovideahighercapacitywhilethelattercanproduce abetterqualitymarkedimage.
Thispaperproposesareversibledatahidingschemebasedon histogrammodification.Itsprincipleistomodifythehistogram constructedbasedontheneighborpixeldifferencesinsteadofthe hostimage’shistogramasin[13].Manypeakpointsexistaround
1434-8411/$–seefrontmatter © 2011 Elsevier GmbH. All rights reserved. doi:10.1016/j.aeue.2011.01.014
Z.Zhaoetal./Int.J.Electron.Commun.(AEÜ)65 (2011) 814–826 815 Original histogram … 0 1 2 … PP+1 Z-1Z … 255 … … … … … 0 1 2 …PP+1 … Z-1Z… 255 … … … 0 1 2 … PP+1 … Z-1Z… 255
“0” “1”
Shifted histogram Histogram with data embedded Fig.1.Principleofreversibledatahidingbasedonhistogrammodification.
thebinzerointhishistogramduetothesimilarityofadjacentpixel values.Besides,manyzeropointsexistinbothsidesofthebinzero. Herethepeakpointreferstotheheightofhistogrambinwiththe largeststatisticalvalue(i.e.,thecountfallinginthe correspond-ingbin),and thezeropointmeansthehistogrambin withzero value.Inourcase,allthedifferencesareclassifiedintolevelsof [−255,255]andeachlevelcorrespondstoahistogrambin.Hence itisreasonabletomodifythehistogramwithamultilevel mecha-nismforhidingmoresecretdata.Indecoder,thehostimagepixels arerecoveredonebyone.Thatis,eachpixelisrecoveredaidedby itspreviouslyrecoveredneighbor.Meanwhile,thesecretdatais extractedfromthemarkedadjacentpixels’differences.
Thispaperisorganizedasfollows.Section2reviewstherelated work.Section 3 describes theproposed schemeincluding data embedding,extractionandimagerecoveryprocedures.Section4 discussesthecapacity estimation, overflowand underflow pre-vention.Experimentalresultsandperformancecomparisonswith otheralgorithmsareshowninSection5.Finally,conclusionsare giveninSection6.
2. Relatedwork
In[13],Nietal.proposedareversibledatahidingmethodbased onhistogrammodification.Inthescheme,partofthecoverimage histogramisshiftedrightwardorleftwardtoproduceredundancy fordataembedding.Theprinciplecanbeillustratedasshownin Fig.1.First,thepeakandzeropointbinsoftheoriginalhistogram arefounddenotedasb(P)andb(Z),respectively.Thenallthebins belongingtob(P)andb(Z)areshiftedrightwardonelevel.Inthis way,thebinofb(P)isemptiedandb(P+1)becomesthenewpeak point.Next,theconfidentialdatacanbeembeddedbymodulating thepixelvaluesequalingP+1.Thatis,ifencounterapixelwith valueequalingP+1,thenonebitconfidentialdatacanbehidden. Forexample,ifthecurrentprocessingconfidentialbitis“0”,we modifythepixelvalueasP;whereasifthecurrentprocessing con-fidentialbitis“1”,thepixelwithvalueP+1iskeptnochanged. Indecoder,thedataextractionandimagerecoveryistheinverse processingofdataembedding.
In[16],Lietal.proposedareversibledatahidingmethodnamed adjacentpixeldifference(APD)basedontheneighborpixel differ-encesmodification.Inthismethod,aninverse“S”orderisadopted toscantheimagepixels.AsshowninFig.2,a3×3imageblockis usedtoillustratethisprinciple.Thescandirectionismarkedasthe blueline,andtheblockcanberearrangedintoapixelsequenceas p1,p2,...,p9.
SupposethehostimageIisan8-bitgraylevelimagesizedas M×N.Thena pixelsequence p1,p2,...,pM×N areobtainedvia
theinverse“S”orderscan.Thedifferencesofadjacentpixelsare computedas: di=
p1 i=1 pi−1−pi 2≤i≤M×N (1)p
1
p
6
p
2
p
5
p
3
p
4
p
7
p
8
p
9
Fig.2. Inverse“S”scanofa3× 3imageblock.
Consideringthepixelvaluessimilaritybetweenpi−1andpi,alarge
quantityofdi(2≤i≤M×N)isequalorcloseto0.Thedifference
histogramisconstructedbasedontheseM×N−1difference statis-tics.Supposethehistogrambinsfromlefttorightaredenotedby b(−255),b(−254),...,b(−1),b(0),b(1),...,b(254),b(255).Fig.3 showsthe512×512Lenaimage’sdifferencehistogram.Obviously mostdifferencesareconcentratedaroundb(0).Whenthecurve spreadsawaytobothsides,itdropsdramatically,andnodifferences fallintothosebinsfarfromb(0).
Basically,APDselectsonepairofbinsb(p1)andb(z1)(suppose
p1<z1)whereb(p1)andb(z1)denotethepeakpointandzeropoint,
respectively.Thenthebinsbetween[b(p1+1),b(z1−1)]areshifted
rightwardonelevel.Thusb(p1+1)areemptiedfordataembedding.
Thatis,ifasecretbit“1”isembedded,thedifferencesequalingp1
areaddedby1.If“0”isembedded,theyarenotchanged.
Toenhancethecapacity,APDcanalsoselecttwopairsof peak-zeropoints,e.g. [b(p1), b(z1)]and [b(z2), b(p2)] (supposep1<z1
-250 -200 -150 -100 -50 0 50 100 150 200 250 0 0.5 1 1.5 2 2.5 3x 10 4
4 0 -4 4 0 -4 4 0 -4 -4 0 4
c
d
Fig.4.HistogrammodificationforEL=0.
andz2<p2).Thenthebinsbetween[b(p1+1),b(z1−1)]areshifted
rightwardone level,andthose between[b(z2+1),b(p2−1)]are
shiftedleftwardonelevel.Thusb(p1+1)andb(p2−1)areemptied
fordataembedding.Thesecretbitsmodulationissimilarasthatin onepairofpeak-zeropointsembedding.Notetherangesof[b(p1),
b(z1)]and[b(z2),b(p2)]mustnotbeoverlapped.
3. Proposedscheme
3.1. Motivation
However,thedisadvantageofAPDmethodisthattheprovided capacityisnotveryhighduetoonlytwopairsofpeak-zeropointsat mostareemployedfordatahiding.Thislimitsitsscopeof applica-tionwherealargequantityofdataistobeembedded.Infact,more pairsofpeak-zeropointscanbeutilized.Motivatedfromthis,this workdesignedamultilevelhistogrammodificationmechanismfor alargecapacitydatahiding.
3.2. Dataembedding
In ourscheme,theinverse“S” order is adoptedtoscan the imagepixelsfordifferencegeneration.Thesecretdataarebinary sequencesproducedbypseudorandomnumbergenerator.Inthe dataembeddingstage,amultilevelhistogrammodificationstrategy isutilized.AnintegerparametercalledembeddinglevelEL(EL≥0) isinvolvedtocontrolthehidingcapacity.AlargerELindicatesmore secretdatacan beembedded. Astheembeddingoperationsfor EL>0aremorecomplicatedthanthoseofEL=0,wedescribethem forEL=0andEL>0separately.
Step1.Inverse“S”scanIintoapixelsequencep1,p2,...,pM×N.
Step2.Computethedifferencesdi(1≤i≤M×N)accordingtoEq.
(1)andconstructahistogrambasedondi(2≤i≤M×N).
Step3.SelectanEL.IfEL=0,executeStep4.IfEL>0,gotoStep5. Step4.DataembeddingforEL=0.
Step4.1.Shifttherightbinsofb(0)rightwardonelevelas:
di=
p1 if i=1 di if di≤0,2≤i≤M×N di+1 if di>0,2≤i≤M×N (2)Step4.2.Examinedi=0(2≤i≤M×N)onebyone.Eachdifference equaling0canbeusedtohideonesecretbit.Ifthecurrent
pro-4 0 -4 4 0 -4 4 0 -4 4 0 -4 4 0 -4 4 0 -4 4 0 -4 4 0 -4
c
d
e
f
g
h
Fig.5.HistogrammodificationforEL=2.
cessingsecretbitw=0,itisnotchanged.Ifw=1,itisaddedby 1.Theoperationisas:
di =
p1 if i=1 di+w if di=0,2≤i≤M×N di if di /=0,2≤i≤M×N (3)The histogram modification strategy for EL=0 is shown in Fig.4(a)–(d)wheretheredandbluearrowsindicateembedding “0”and“1”,respectively.Afterthat,gotoStep6.(For interpreta-tionofthereferencestocolourinthetext,thereaderisreferred tothewebversionofthearticle.)
Step5.DataembeddingforEL>0.
Step5.1.Shifttherightbinsofb(EL)rightwardEL+1levels,and shifttheleftbinsofb(−EL)leftwardELlevelsas:
di=
⎧
⎪
⎨
⎪
⎩
p1 if i=1 di if −EL≤di≤EL,2≤i≤M×N di+EL+1 if di>EL,2≤i≤M×N di−EL if di<−EL,2≤i≤M×N (4)Step5.2.Examinedi=0(2≤i≤M×N)intherangeof[−EL,EL] onebyone.Themultileveldataembeddingstrategyisdescribed asfollows.
Step5.2.1.Embedthesecretdataas:
di =
⎧
⎪
⎨
⎪
⎩
p1 if i=1 di if −EL<di<EL,2≤i≤M×N 2×EL+w if di=EL,2≤i≤M×N −2×EL−w+1 if di=−EL,2≤i≤M×N (5) Step5.2.2.ELisdecreasedby1.Z.Zhaoetal./Int.J.Electron.Commun.(AEÜ)65 (2011) 814–826 817
Fig.6. ExampleofdataembeddingforEL=0.
Fig.7. ExampleofdataextractionandimagerecoveryforEL=0.
Fig.23.MarkedCarimagesobtainedbyourscheme(left:220801bitshidden,33.05dB)andKimetal.’sscheme[17](right:171244bitshidden,31.40dB)withEL=9.
andAPD2denoteoneandtwopairsofpeak-zeropointsareused
respectively.ThecomparisonresultsarealsogiveninTable1.Our schemecanprovideahighercapacitythanLietal.’smethodwith goodmarkedimagesquality.
Next,ourschemeiscomparedwithKimetal.’smethod[17]. Althoughboth are based onmultilevel histogram modification, the histogram construction mechanisms are different. In gen-eral,thecapacityof differencehistogram modificationis jointly affectedby the total number of differences and the their con-centricity tob(0). In [17], the differences are computed based onsubimages’correlation andhence thenumberof differences is determined by the number of subimages. For example, if a 512×512hostimageissubsampledinto16 equal-sized subim-ages,thereare512×512×15/16=245760differencesproduced. Incontrast,thereare512×512−1=262143differencesproduced inourscheme.The histogrambinsbelongingto[b(−30),b(30)] obtainedby [17] and our schemeare shown in Fig. 11. Obvi-ously,moredifferencesinourhistogramsareconcentratedaround b(0).Asaresult,alargercapacitycanbeprovidedinourscheme thanin[17].Moreover,b(−1)isemptiedafterembeddingin[17] for the shifting leftward is one level farther than that in our scheme,and consequently theintroduced distortions are more serious.
Theperformancecomparisonsofourscheme(markedasblue) andKimetal.’smethod(markedasred)areshowninFigs.12–17. The horizontal axisdenotes the EL setfrom 0 to 9. The verti-calaxesofcapacityandPSNRarenormalizedas[0,1.0]bpp(bit perpixel)and[30,55]dB,respectively.Intheseexperiments,the hostimagesarepartitionedinto16equal-sizedsubimagesin[17]. ThesixmarkedimagesobtainedbyourschemeandKimetal.’s schemeareshowninFigs. 18–23.Alltheseresultsdemonstrate not only the capacities but also the PSNRs in our method are improved.Inotherwords,eventhoughmoresecretdataembedded inourscheme,themarkedimagesqualityisstillbetterthanthose in[17].
6. Conclusions
Areversibledatahidingschemeisproposedinthispaper.The multilevelhistogram modificationisemployedfor data embed-ding.Ononehand,ahighercapacityisprovidedcomparedwithone ortwolevelhistogrammodificationbasedmethods.Ontheother hands,assecretdataisembeddedindifferencesofadjacentpixels values,themarkedimagesqualityisimprovedcomparedwiththat inpreviousmultilevelhistogrammodificationbasedwork.
Acknowledgments
Theauthorsthanktheeditorandtheanonymousreviewersfor theirconstructivecommentsandvaluablesuggestionsfor readabil-ityimprovement.ThisworkisfinanciallysupportedbytheNational ScientificFundofChina(Nos.61003255and61071128).
References
[1]Pan JS, Huang HC, Jain LC. Information hiding and applications. Berlin-Heidelberg,Germany:Springer;2009.
[2]CheddadA,CondellJ,CurranK,KevittsPM.Digitalimagesteganography:survey andanalysisofcurrentmethods.SignalProcessing2010;90(3):727–52. [3]TianJ.Reversibledataembeddingusingadifferenceexpansion.IEEE
Transac-tionsonCircuitsandSystemsforVideoTechnology2003;13(8):890–6. [4]Alattar AM. Reversible watermark using the difference expansion of a
generalized integer transform. IEEE Transactions on Image Processing 2004;13(8):1147–56.
[5]WengS,ZhaoY,PanJS.Anovelreversibledatahidingscheme.International JournalofInnovativeComputing,InformationandControl2008;4(2):351–8. [6] ChangCC,LuTC.Adifferenceexpansionorienteddatahidingschemefor
restoringthe originalhost images.TheJournalof SystemsandSoftware 2006;79(12):1754–66.
[7]LeeCC,WuHC,TsaiCS,ChuYP.Adaptivelosslesssteganographicschemewith centralizeddifferenceexpansion.PatternRecognition2008;41(6):2097–106. [8]LinCC,HsuehNL.Alosslessdatahidingschemebasedonthree-pixelblock
differences.PatternRecognition2008;41(4):1415–25.
[9]LuTC,ChangCC.Losslessnibbleddataembeddingschemebasedondifference expansion.ImageandVisionComputing2008;26(5):632–8.
[10]HsiaoJY,ChanKF,ChangJM.Block-basedreversibledataembedding.Signal Processing2009;89(4):556–69.
[11]LouDC,HuMC,LiuJL.Multiplelayerdatahidingschemeformedicalimages. ComputerStandards&Interfaces2009;31(2):329–35.
[12]WengS,ZhaoY,PanJS,NiR.Reversibledatahidingusingthe compand-ing techniqueand improvedDEmethod. CircuitsSystemsSignalProcess 2008;27(2):229–45.
[13]NiZ,ShiY,AnsariN,SuW.Reversibledatahiding.In:IEEEproceedingsof ISCAS’03,vol.2.2003.p.II-912–915.
[14]HuangHC,FangWC,TsaiIT.Reversibledatahidingusinghistogram-based differenceexpansion.IEEEInternationalSymposiumonCircuitsandSystems 2009:1661–4.
[15]TsaiP,HuYC,YehHL.Reversibleimagehidingschemeusingpredictivecoding andhistogramshifting.SignalProcessing2009;89(6):1129–43.
[16]Li YC, Yeh CM,ChangCC. Data hiding basedonthe similaritybetween neighboring pixels with reversibility. Digital Signal Processing 2009, doi:10.1016/j.dsp.2009.10.025.
[17]KimKS,LeeMJ,LeeHY,LeeHK.Reversibledatahidingexploitingspatial correla-tionbetweensub-sampledimages.PatternRecognition2009;42(11):3083–96. [18]LinCC,TaiWL,ChangCC.Multilevelreversibledatahidingbasedonhistogram modificationofdifferenceimages.PatternRecognition2008;41(12):3582–91. [19]GaoX,AnL,LiX,TaoD.Reversibilityimprovedlosslessdatahiding.Signal
Processing2009;89(10):2053–65.
[20]TaiWL,YehCM,ChangCC.Reversibledatahidingbasedonhistogram modifi-cationofpixeldifferences.IEEETransactionsonCircuitsandSystemsforVideo Technology2009;19(6):906–10.
[21] YangB,SchmuckerM,FunkW,BrushC,SunS.IntegerDCT-basedreversible watermarking for images using companding technique. Proceedings of
826 Z.Zhaoetal./Int.J.Electron.Commun.(AEÜ)65 (2011) 814–826
SPIE, Security,Steganography andWatermarkingof MultimediaContents 2004;5306:405–15.
[22]YangB,SchmuckerM,NiuX,BuschC,SunSH.Integer-DCT-basedreversible image watermarkingby adaptive coefficientmodification.Proceedings of SPIE,Security,Steganography,andWatermarkingofMultimediaContents 2005;5681:218–29.
[23]XuanG,ShiYQ,YaoQ,NiZ,YangC,GaoJ,etal.Losslessdatahidingusing histogramshiftingmethodbasedonintegerwavelets.International Work-shoponDigitalWatermarking,LectureNotesinComputerScience2006;4283: 323–32.
[24]LuZM,WangJX,LiuBB.Animprovedlosslessdatahidingschemebased onimageVQ-indexresidualvaluecoding.JournalofSystemsandSoftware 2009;82(6):1016–24.
[25] WangJX,LuZM.ApathoptionallosslessdatahidingschemebasedonVQindex tablejointneighboringcoding.InformationSciences2009;179(19):3332–48.
[26]YangB,LuZM,SunSH.ReversiblewatermarkingintheVQ-compresseddomain. In:Proceedingsofthefifthinternationalconferenceonvisualization,imaging, andimageprocessing.2005.p.289–303.
[27]ChangCC,LinCY,FanYH.Losslessdatahidingforcolorimagesbasedonblock truncationcoding.PatternRecognition2008;41(7):2347–57.
[28]MobasseriBG,CinalliD.Losslesswatermarkingofcompressedmediausing reversiblydecodablepackets.SignalProcessing2006;86(5):951–61. [29] LiLD,GuoBL.Localizedimagewatermarkinginspatialdomainresistantto
geometricattacks.InternationalJournalofElectronicsandCommunications (AEU)2009;63(2):4.p.123–31.
[30]WangZH,KieuTD,ChangCC,LiMC.Anovelinformationconcealingmethod basedonexploitingmodificationdirection.JournalofInformationHidingand MultimediaSignalProcessing2010;1(1):1–9.
[31] BesdokE.Hidinginformationinmultispectralspatialimages.International JournalofElectronicsandCommunications(AEU)2005;59:15–24.