关闭

关闭

封号提示

内容

首页 Reversible Watermarking Techniques An Overview …

Reversible Watermarking Techniques An Overview and a Classification.pdf

Reversible Watermarking Techniq…

上传者: hiyyt 2011-12-26 评分 5 0 155 21 706 暂无简介 简介 举报

简介:本文档为《Reversible Watermarking Techniques An Overview and a Classificationpdf》,可适用于IT/计算机领域,主题内容包含HindawiPublishingCorporationEURASIPJournalonInformationSecurityVolume,Arti符等。

HindawiPublishingCorporationEURASIPJournalonInformationSecurityVolume,ArticleID,pagesdoi:ReviewArticleReversibleWatermarkingTechniques:AnOverviewandaClassificationRobertoCaldelli,FrancescoFilippini,andRudyBecarelliMICC,UniversityofFlorence,VialeMorgagni,Florence,ItalyCorrespondenceshouldbeaddressedtoRobertoCaldelli,robertocaldelliunifiitReceivedDecemberAcceptedMayAcademicEditor:JiwuWHuangCopyrightRobertoCaldellietalThisisanopenaccessarticledistributedundertheCreativeCommonsAttributionLicense,whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycitedAnoverviewofreversiblewatermarkingtechniquesappearedinliteratureduringthelastfiveyearsapproximatelyispresentedinthispaperInadditiontothisageneralclassificationofalgorithmsonthebasisoftheircharacteristicsandoftheembeddingdomainisgiveninordertoprovideastructuredpresentationsimplyaccessibleforaninterestedreaderAlgorithmsaresetinacategoryanddiscussedtryingtosupplythemaininformationregardingembeddinganddecodingproceduresBasicconsiderationsonachievedresultsaremadeaswellIntroductionDigitalwatermarkingtechniqueshavebeenindicatedsofarasapossiblesolutionwhen,inaspecificapplicationscenario(authentication,copyrightprotection,fingerprinting,etc),thereistheneedtoembedaninformativemessageinadigitaldocumentinanimperceptiblewaySuchagoalisbasicallyachievedbyperformingaslightmodificationtotheoriginaldatatryingto,atthesametime,satisfyotherbindingssuchascapacityandrobustnessWhatisimportanttohighlight,beyondthewayalltheseissuesareachieved,itisthatthis“slightmodification”isirreversible:thewatermarkedcontentisdifferentfromtheoriginaloneThismeansthatanysuccessiveassertion,usage,andevaluationmusthappenona,thoughweakly,corruptedversion,iforiginaldatahavenotbeenstoredandarenotreadilyavailableItisnowclearthatindependenceoftheapplicationscenario,thiscannotalwaysbeacceptableUsuallywhendealingwithsensitiveimagerysuchasdeepspaceexploration,militaryinvestigation,andrecognition,andmedicaldiagnosis,theendusercannottoleratetorisktogetadistortedinformationfromwhatheiswatchingatOneexampleaboveall:aradiologistwhoischeckingaradiographicimagetoestablishifacertainpathologyispresentornotItcannotbeacceptedthathisdiagnosisiswrongboth,firstly,tosafeguardthepatient’shealthand,secondly,toprotecttheworkoftheradiologisthimselfInsuchcases,irreversiblewatermarkingalgorithmsclearlyappearnottobefeasibleduetothisstrictrequirement,anothercategoryofwatermarkingtechniqueshavebeenintroducedinliteraturewhicharecataloguedasreversible,where,withthisterm,itistobeintendedthattheoriginalcontent,otherthanthewatermarksignal,isrecoveredfromthewatermarkeddocumentsuchthatanyevaluationcanbeperformedontheunmodifieddataThusdoing,thewatermarkingprocessiszeroimpactbutallows,atthesametime,toconveyaninformativemessageReversiblewatermarkingtechniquesarealsonamedasinvertibleorlosslessandwereborntobeappliedmainlyinscenarioswheretheauthenticityofadigitalimagehastobegrantedandtheoriginalcontentisperemptorilyneededatthedecodingsideItisimportanttopointoutthat,initially,ahighperceptualqualityofthewatermarkedimagewasnotarequirementduetothefactthattheoriginalonewasrecoverableandsimpleproblemsofoverflowandunderflowcausedbythewatermarkingprocesswerenottakenintoaccounttooSuccessivelyalso,thisaspecthasbeenconsideredasbasictopermittotheendusertooperateonthewatermarkedimageandtopossiblydecidetoresorttotheuncorruptedversioninasecondtimeifneededEURASIPJournalonInformationSecuritySemifragileRobustFragileReversibleFigure:CategorizationofreversiblewatermarkingtechniquesReversiblealgorithmscanbesubdividedintotwomaincategories,asevidencedinFigure:fragileandsemifragileMostofthedevelopedtechniquesbelongtothefamilyoffragilethatmeansthattheinsertedwatermarkdisappearswhenamodificationhasoccurredtothewatermarkedimage,thusrevealingthatdataintegrityhasbeencompromisedAninferiornumber,inpercentage,aregroupedinthesecondcategoryofsemifragilewherewiththistermitisintendedthatthewatermarkisabletosurvivetoapossibleunintentionalprocesstheimagemayundergo,forinstance,aslightJPEGcompressionSuchfeaturecouldbeinterestinginapplicationswhereacertaindegreeoflossycompressionhastobetoleratedthatis,theimagehastobedeclaredasauthenticevenifslightlycompressedWithinthislastcategorycanalsobeincludedarestrictedsetoftechniquesthatcanbedefinedasrobustwhichareabletocopewithintentionalattackssuchasfiltering,partialcropping,JPEGcompressionwithrelativelylowqualityfactors,andsoonTherationalebehindthispaperistoprovideanoverview,ascompleteaspossible,andaclassificationofreversiblewatermarkingtechniques,whiletryingtofocusontheirmainfeaturesinamannertoprovidetothereadersbasicinformationtounderstandifacertainalgorithmmatcheswithwhattheywerelookingforInparticular,ourattentionhasbeendedicatedtopapersappearedapproximatelyfromyearstillinfact,duetothehugeamountofworksinthisfield,wehavedecidedtorestrictourwatchtothelastimportanttechniquesAnywaywecouldnotforgetsome“old”techniquesthatareconsideredasreferencethroughoutthepaper,suchas–,thoughtheyarenotdiscussedindetailThepapertriestocategorizethesetechniquesaccordingtotheclassificationpicturedinFigureandbyaddinganinterestingdistinctionregardingtheembeddingdomaintheyworkon:spatialdomain(pixel)ortransformeddomain(DFT,DWT,etc)Thepaperisstructuredasfollows:inSection,fragilealgorithmsareintroducedandsubdividedintotwosubclassesonthebasisoftheadopteddomaininSection,techniqueswhichprovidefeaturesofsemifragilenessandorrobustnessarepresentedandclassifiedagainaccordingtothewatermarkingdomainSectionconcludesthepaperFragileAlgorithmsFragilealgorithmscoverthemajorityofthepublishedworksinthefieldofreversibleWiththetermfragileawatermarkingtechniquewhichembedsacodeinanimagethatisnotreadableanymoreifthecontentisalteredConsequentlytheoriginaldataarenotrecoverabletooSpatialDomainThissubsectionisdedicatedtopresentsomeofthemainworksimplementingfragilereversiblewatermarkingbyoperatinginthespatialdomainOneofthemostimportantworksinsuchafieldhasbeenpresentedbyTian,Itpresentsahighcapacity,highvisualquality,andreversibledataembeddingmethodforgrayscaledigitalimagesThismethodcalculatesthedifferenceofneighboringpixelvaluesandthenselectssomeofsuchdifferencestoperformadifferenceexpansion(DE)Insuchdifferentvalues,apayloadBmadebythefollowingpartswillbeembedded:(i)aJBIGcompressedlocationmap,(ii)theoriginalLSBvalues,and(iii)thenetauthenticationpayloadwhichcontainsanimagehashToembedthepayload,theprocedurestartstodefinetwoamounts,theaveragelandthedifferenceh(see())Givenapairofpixelvalues(x,y)inagrayscaleimage,withx,yZ,x,y,l=xyh=xy,()andgivenlandh,theinversetransformcanberespectivelycomputedaccordingto()x=lhy=lh()ThemethoddefinesdifferentkindsofpixelcouplesaccordingtothecharacteristicsofthecorrespondinghandbehavesslightlydifferentforeachofthemduringembeddingTwoarethemaincategories:changeableandexpandabledifferences,letusseebelowfortheirdefinitions,respectivelyDefinitionForagrayscalevaluedpair(x,y)adifferencenumberhischangeableifhbmin((l),l)()DefinitionForagrayscalevaluedpair(x,y)adifferencenumberhisexpandableif|hb|min((l),l)()Thisisimposedtopreventoverflowunderflowproblemsforthewatermarkedpixels(x′,y′)Toembedabitb=(,)ofthepayload,itisnecessarytomodifytheamounthobtainingh′whichiscalledDEEURASIPJournalonInformationSecurityTable:PayloadsizeversusPSNRofLenaimagePayloadSize(bits)BitRate(bpp)PSNR(dB)(DifferenceExpansion)accordingto()forexpandabledifferencesh′=hb,b=LSB(h′),()and()forchangeableonesh′=hb,b=LSB(h′),()byreplacinghwithh′within(),thewatermarkedpixelvaluesx′andy′aregotThebasicfeaturewhichdistinguishesexpandabledifferencesfromchangeableonesisthatthefirstonescancarryabitwithoutaskingforsavingtheoriginalLSBThatyieldstoareducedtotalpayloadBAlocationmaptakesintoaccountofthediversedisjointcategoriesofdifferencesToextracttheembeddeddataandrecovertheoriginalvalues,thedecoderusesthesamepatternadoptedduringembeddingandapplies()toeachpairThentwosetsofdifferencesarecreated:CforchangeablehandNCfornotchangeablehBytakingallLSBsofdifferencesbelongingtoCset,abitstreamBiscreatedFirstly,thelocationmapisrecoveredandusedtogetherwithBtorestoretheoriginalhvaluessecondly,byusing()theoriginalimageisobtained,lastly,theembeddedpayload(theremainingpartofB)isusedforauthenticationcheckbyresortingtotheembeddedhashTianappliesthealgorithmto“Lena”(),bppgrayscaleimageTheexperimentalresultsareshowninTable,wheretheembeddedpayloadsize,thecorrespondingbitrate,andPSNRsofthewatermarkedimagearelistedAsDEincreases,thewatermarkhastheeffectsimilartomildsharpeninginthemidtoneregionsApplyingtheDEmethodon“Lena,”theexperimentalresultsshowthatthecapacityversusdistortionisbetterincomparisonwiththeGLSBmethodproposedin,andtheRSmethodproposedinThepreviousmethodhasbeentakenandextendedbyAlattarinInsteadofusingdifferenceexpansionappliedtopairsofpixelstoembedonebit,inthiscasedifferenceexpansioniscomputedonspatialandcrossspectraltripletsofpixelsinordertoincreasehidingcapacitythealgorithmembedstwobitsineachtripletWiththetermtripletavectorcontainingthepixelvaluesofacoloredimageisintendedinparticular,therearetwokindsoftriplets(i)SpatialTriplet:threepixelvaluesoftheimagechosenfromthesamecolorcomponentwithintheimageaccordingtoapredeterminedorder(ii)CrossspectralTriplet:threepixelvaluesoftheimagechosenfromdifferentcolorcomponents(RGB)Theforwardtransformforthetriplett=(u,u,u)isdefinedasv=uwuuN,v=uu,v=uu,()whereNandwareconstantForspatialtriplets,N=andw=,whileincrossspectraltriplets,N=andw=Ontheotherside,theinversetransform,f(),forthetransformedtripletst′=(v,v,v)isdefinedasu=vvvN,u=vu,u=vu()Thevaluevandvareconsideredforwatermarkingaccordingto()v′=vb,v′=vb,()foralltheexpandabletriplets,whereexpandablemeansthat(v′v′)satisfiesalimitationsimilarlytowhathasbeenproposedinthepreviouspapertoavoidoverflowunderflowIncaseofonlychangeabletriplets,v′=vb(v′changescorrespondingly),butthesameboundforthesumofthesetwoamountshastobeverifiedagainAccordingtotheabovedefinition,thealgorithmclassifiesthetripletsinthefollowinggroups()S:containsallexpandabletripletswhosevTandvT(T,Tpredefinedthreshold)()S:containsallchangeabletripletsthatarenotinS()S:containsthenotchangeabletriplets()S=SScontainsallchangeabletripletsIntheembeddingprocess,thetripletsaretransformedusing()andthendividedintoS,SandSS,andSaretransformedinSwandSw(watermarked)andthepixelvaluesoftheoriginalimageI(i,j,k)arereplacedwiththecorrespondingwatermarkedtripletsinSwandSwtoproducethewatermarkedimageIw(i,j,andk)ThealgorithmusesabinaryJBIGcompressedlocationmapM,toidentifythelocationofthetripletsinS,S,andSwhichbecomespartofthepayloadtogetherwiththeLSBofchangeabletripletsInthereadingandrestoringprocess,thesystemsimplyfollowstheinversestepsoftheencodingphaseAlattarEURASIPJournalonInformationSecurityTable:EmbeddedpayloadsizeversusPSNRforcoloredimagesLenaBaboonFruitsPayload(bits)PSNR(dB)Payload(bits)PSNR(dB)Payload(bits)PSNR(dB),,,,,,,,,,,,,,,,,,,,,Table:ComparisonresultsbetweenTian’sandAlattar’salgorithmGrayscaleLenaGrayscaleBarbaraTian’sAlgAlattar’sAlgTian’sAlgAlattar’sAlgPSNR(dB)Payload(bits)Payload(bits)PSNR(dB)Payload(bits)Payload(bits)whQuadq=(u,u,u,u)Figure:QuadsconfigurationinanimagetestedthealgorithmwiththreeRGBimages,Lena,Baboon,andFruitsThealgorithmisappliedrecursivelytocolumnsandrowsofeachcolorcomponentThewatermarkisgeneratedbyarandombinarysequenceandT=TinallexperimentsInTable,PSNRsofthewatermarkedimagesareshownIngeneral,thequalitylevelisaboutdBwithabitrateofbitscoloredpixelInTable,itisreportedalsotheperformancecomparisonintermsofcapacitybetweentheTian’salgorithmandthisone,byusinggrayscaleimagesLenaandBarbaraFromtheresultsofTable,thealgorithmproposedoutperformstheTian’stechniqueatlowerPSNRsAthigherPSNRsinstead,theTian’smethodoutperformstheproposedAlattarproposedinanextensionofsuchatechnique,tohidetripletsofbitsinthedifferenceexpansionofquadsofadjacentpixelsWiththetermquadsavectorcontainingthepixelvalues(adjacentpixelvalues)fromdifferentlocationswithinthesamecolorcomponentoftheimageisintended(seeFigure)Thedifferenceexpansiontransform,f(),forthequadq=(u,u,u,u)isdefinedasin()v=auauauauaaaa,v=uu,v=uu,v=uu()Theinversedifferenceexpansiontransform,f(),forthetransformedquadq′=(v,v,v,v)iscorrespondinglydefinedasin()u=v(aaa)v(aa)vavaaaa,u=vu,u=vu,u=vu()Similarlytotheapproachpreviouslyadopted,quadsarecategorizedinexpandableorchangeableanddifferentlytreatedduringwatermarkingthentheyaregroupedasfollows()S:containsallexpandablequadswhosevT,vT,vTwithv,v,vtransformedvaluesandT,T,andTpredefinedthreshold()S:containsallchangeablequadsthatarenotinS()S:containstherestofquads(notchangeable)()S:containsallchangeablequads(S=SS)EURASIPJournalonInformationSecurityIntheembeddingprocessthequadsaretransformedbyusing()andthendividedintothesetsS,S,andSSandSaremodifiedinSwandSw(thewatermarkedversions)andthepixelvaluesoftheoriginalimageI(i,j,andk)arereplacedwiththecorrespondingwatermarkedquadsinSwandSwtoproducethewatermarkedimageIw(i,j,k)WatermarkextractionandrestoringprocessproceedsinverselyasusualInthepresentedexperimentalresults,thealgorithmisappliedtoeachcolorcomponentofthreeRGBimages,Baboon,Lena,andFruitssettingT=T=TinallexperimentsTheembeddingcapacitydependsonthenatureoftheimageitselfInthiscase,theimageswithalotoflowfrequenciescontentsandhighcorrelation,likeLenaandFruits,producemoreexpandabletripletswithlowerdistortionthanhighfrequencyimagessuchasBaboonInparticularwithFruits,thealgorithmisabletoembedkbitswithaPSNRdB,butwithonlykbitsimagequalityincreasesatdBItisinterestingtoverifythatwithBaboonthealgorithmisabletoembedkbitsorkbitsachievingaPSNRofdBandofdB,respectivelyTheproposedmethodiscomparedwithTian’salgorithm,usinggrayscaleimages,LenaandBarbaraAtPSNRhigherthandB,quadbasedtechniqueoutperformsTian,whileatlowerPSNRTianoutperforms(marginally)theproposedtechniquesThequadbasedalgorithmisalsocomparedwithmethodusinggrayscaleimageslikeLenaandBarbaraAlso,inthiscasetheproposedmethodoutperformsCelikatalmostallPSNRsTheproposedalgorithmisalsocomparedwiththepreviousworkofAlattardescribedinTheresultsrevealthattheachievablepayloadsizeforthequadbasedalgorithmisabout,bitshigherthanforthespatialtripletsbasedalgorithmatthesamePSNRfurthermore,thePSNRisaboutdBhigherforthequadbasedalgorithmthanforthespatialtripletbasedalgorithmatthesamepayloadsizeFinally,in,Alattarhasproposedafurthergeneralizationofhisalgorithm,byusingdifferenceexpansionofvectorscomposedbyadjacentpixelsThisnewmethodincreasesthehidingcapacityandthecomputationefficiencyandallowstoembedintotheimageseveralbits,ineveryvector,inasinglepassAvectorisdefinedasu=(u,u,,uN),whereNisthenumberofpixelvalueschosenfromNdifferentlocationswithinthesamecolorcomponent,taken,accordingtoasecretkey,fromapixelsetofabsizeInthiscase,theforwarddifferenceexpansiontransform,f(),forthevectoru=(u,u,,uN)isdefinedasv=Ni=aiuiNi=ai,v=uu,vN=uNu,()whereaiisaconstantinteger,ah,bwandab=,(wandharetheimagewidthandheight,resp)Theinversedifferenceexpansiontransform,f(),forthetransformedvectorv=(v,v,,vN),isdefinedasu=vNi=aiviNi=ai,u=vu,uN=vNu()Similarlytowhatwasdonebefore,thevectoru=(u,u,,uN)canbedefinedexpandableif,forall(b,b,,bN),,v=f(u)canbemodifiedtoproducev=(v,v,,vN)withoutcausingoverflowand

精彩专题

职业精品

上传我的资料

热门资料

资料评价:

/ 19
所需积分:0 立即下载

意见
反馈

返回
顶部

Q