会员注册 | 登录 | 微信快捷登录 支付宝快捷登录 QQ登录 微博登录 | 帮助中心 人人文库renrendoc.com美如初恋!
站内搜索 百度文库

热门搜索: 直缝焊接机 矿井提升机 循环球式转向器图纸 机器人手爪发展史 管道机器人dwg 动平衡试验台设计

   首页 人人文库网 > 资源分类 > PDF文档下载

Passive detection of copy-paste forgery between JPEG images

  • 资源星级:
  • 资源大小:2.21MB   全文页数:13页
  • 资源格式: PDF        下载权限:注册会员/VIP会员
您还没有登陆,请先登录。登陆后即可下载此文档。
  合作网站登录: 微信快捷登录 支付宝快捷登录   QQ登录   微博登录
友情提示
2:本站资源不支持迅雷下载,请使用浏览器直接下载(不支持QQ浏览器)
3:本站资源下载后的文档和图纸-无水印,预览文档经过压缩,下载后原文更清晰   

Passive detection of copy-paste forgery between JPEG images

J.Cent.SouthUniv.20121928392851DoIl0.1007/sl177l一012一l3505PassivedetectionofcopypasteforgerybetweenJPEGimagesLIXianghua香7匕ZttAOYuqian0前,LIAOMiao,F.Y_Shih,YQ,Shi垒Springer1.SchoolofCivilEnginecring,CentralSouthUniversity,Changsha410083,China2.SchoolofGeoscienccsandlnfoPhysics,CentralSouthUniversity,Changsha410083,China3.ComputingSciences,NewJerseyInstituteofTechnology,NewarkNJ07102,USA4.ElectricalandComputerEngineering,NewJerseyInstituteofTechnology,NewarkNJ07102,USA◎CentralSouthUniversityPressandSpringerVerlagBerlinHeidelberg2012AbstractAblinddigita1jmageforensicmethodf0rdetectingcopypasteforgerybetweenJPEGimageswasproposed.I、vocopypastetamperingscenarioswereintroducedatfirstthetamperedimagewassavedinanuncompressedformatorinaJPEGcompressedformat.Thentheproposeddetectionmethodwasanalyzedandsimulatedforal1thecasesofthetwotamperingscenarios.ThetamperedregionisdetectedbycomputingtheaveragedSHillofabsolutedifierenceASADimagesbetweentheexaminedimageandaresavedJPEGcompressedimageatdifferentqualityfactors.Theexperimentalresultsshowtheadvantagesoftheproposedmethodcapabilityofdetectingsinai1and/ormultipletamperedregions.simplecomputation.andhencefastspeedinprocessing.KeywordsimageforensicJPEGcompressioncopypasteforgerypassivedetectiontamperedimagecompressedimage1IntroductionTheauthenticityofdigitalimagesplaysanimportantroleinmanyfieldssuchasforensicinvestigation,criminalinvestigation,surveillancesystems,intelligenceservices,disasterforecast,andjournalism1.Theartofmakinganimagefakeryhasalonghistory.Intherapidlydevelopingdigitalage,theimagecanbetamperedeasilywithoutleavingperceptibletraces.Undoubtedly,thisbringsdifficultyforpeopletoprovethetrustworthinessofdigitalimages.Therefore,theimageforensicstechnologiesarebecomingincreasinglyimportant23.Recently,manypassiveimageforensicauthenticitytechniqueshavebeendevelopedwhichcanbegroupedintothefollowingfivecategories【41pixelbasedtechniques,2format.basedtechniques,3camerabasedtechniques,4physicallybasedtecbniques,and5geometricbasedtechniques.Thesetechniquesareusedtoidentifytheimagesourcesandexposeforgeriesthroughdetectingsomeintrinsicimageregularitiesorsomecommontamperinganomaliesf5.JPEGisacommonlyusedcompressionstandardforphotographicimages.Thedegreeofcompressioncanbeadjusted,allowingaselectabletradeoffbetweenstoragesizeandimagequality.JPEGisalsothemostcommonimageformat,usedbydigitalcamerasandotherphotographicimagecapturedevices.JPEGcompressioniSbasedonblocksplittinganddiscretecosinetransforillDCT,soitintroducesblockingartifact.InfacttheblockingartifactcanberegardedasaspeciaIhiddenwatermarkinJPEGimages.andtamperingcanbefoundbydetectingtheintegrityandconsistenceofthespecialwatermark.InRef.6,theblockingartifactgridBAGextractionmethodwasproposedtodetectthedoctoredJPEGimagesbycheckingthemismatchedblockartifactcausedbycopypastedJPEGimages.InRe7,LU0etalpresentedtheblockingartifactcharacteristicsmatrixBACM,whichexhibitsregularsymmeica1shapefororiginalJPEGimages,andappliedittoexposedigitalforgeriesbydetectingthesymmetricalchangeofBACM.ThemethodisonlyefHcientforthecroppedandrecompressedimages.BasedonBACM,BARNIetal『81proposedtwoalgorithmsofdetectingthecopypastetamperingbydoubleJPEGdetectionandimagesegmentation.ButbOthofthealgorithinsaretimeconsuming,andthesegmentationcouldbeverydjfficultforsometamperedimagesifthetamperedregionscannotbesegmentedSuccessfullv.11omeasuretheinconsistenciesofblockingartifact,YEeta1f91presentedafastquantizationtableestimationalgorithmbasedonthehistogrampowerspectrumofDCTcoefficients.Butsuspiciousareamustbefirstlyselectedforevaluation,whichisactuallydifficult.InReff101,astatistica1modelbasedonthegeneralizedBenfordslawFoundationitemProject61172184suppolledbytheNationalNaturalScienceFoundationofChinaProject200902482supposedbyChinaPostdoctoralScienceFoundationSpeciallyFundedProjectProject12JJ6062supportedbytheNaturalScienceFoundationofHunanProvince,ChinaReceiveddate20120202Accepteddate20120502CorrespondingauthorZHAOYuqian,AssociateProfessor,PhDTel十8689929876Emailzhaocsuhotmailcom2840一兰三二三forgerydetectionmethodisproposed.fortheprobabilitydistributionsofthefirstdigitsofthequantizedJPEGcoefficientsisusedtodetecttheJPEGcompressionandimageforensics.LINetal『1l1developedamethodfordetectingchetamperedJPEGimagesbyexaminingthedoublequantizationeffecthiddenamongtheDCTcoefficients,andcomputingtheblockposteriorprobabilitymapBPPMbyBayesianapproach.CopymoveforgerydetectionmethodswereproposedinRefs.12一l6】,buttheyareonlyeffectivewhenapartoftheimageiscopiedandpastedintoanotherpartofthesameimage.LUOetal17introducedtheJPEGquantization,rounding,andtruncationerrorstoidentif3,whetherabitmapimagehaspreviouslybeenJPEGcompressed.toestimatethequantizationstepsofaJPEGimage,andtodetectthequantizationtableofaJPEGimage.WANGetal18】proposedatamperedregionlocalizationmethodbyemployingprincipalcomponentanalysisPCAonJPEGcompressionnoise,inwhichtheextractedhighfrequencyquantizationnoiseregioninthedifferenceimagebetweentheorigina1imageandJPEGcompressedimageisconsideredasthetamperedregion.Fordigitalforensics,FARIDetalhaveconductedsomepioneeringworksandhavedevelopedsomeeffectivemethodsf192l】.InRef.『2o1,JPEGghostwasproposedtodetectimagetampering.ThisisacomputationallysimpletechniquefordetectingtamperedregionwhenaJPEGcompressedimagewithqualityfactorQIisinsertedintoanuncompressedimage,andthenthetamperedimageissavedataJPEGcompressedimagewithaqualityfactorQ2andQ2Q1.InspiredbyRe£201,weproposeinthisworkablinddigitalimageforensicsmethodfordetectingcopypasteforgerybetweenJPEGimagesItworksfortwotamperingscenarios.First,tamperingisconductedbetweenJPEGimages,andthecompositeimageissavedinanuncompressedformat,suchasBMPorTIForPNGSecond,tamperingisdonebetweenJPEGimages.andthecompositeimageissavedintheJPEGformat.TheregionsthatareobviouslybrighterordarkerthantherestintheimageofaveragedsumofabsolutedifferenceASADareidentifiedasthetamperedregions.AlthoughtheJPEGqualityfactorisrelevantinourproposedmethod,theestimationoftheprimaryquantizationmatrix10,2224isnotrequired.2MethodologyInthissection,wefirstintroducetwocopypasteimagetamperingscenariosandbrieflyreviewJPEGcompressionanddecompression.Thenweanalyzetwotamperingscenariosbysimulationanddetectthetamperedimagesbyexamples.Finally,thecopypaste2.1TamperingscenariosCopy。pasteprocessingisarudimentaryandoftenprimarymethodfordigitalimagetampering.AsJPEGisawidelyusedimageformatinourdailylife.wediscusscopy。pastetamperingbetweenJPEGimageswhichhavedifferentJPEGqualityfactors.Thatis,aregioncanbeselectedcomingfromanJPEGcompressedimagewithaqualityfactorofQ1A.Itisthendecompressedand1nsertedintoanotherimagewiththeJPEGqualityfactorofQ1B.ThecompositeimageisthensavedaseitheranuncompressedformatscenarioI、oraJPEGcompressedformatwithaqualityfactorofQ。scenarioIi.Ifitisnotspecified,intherestofthepaper,wedenoteJPEGqualityfactorbycapitalletterQ,andquantizationstepbylowercaseletterq.NotethattheIocationsofcopyandpastecanbearbitraryandtheabovecopypasteprocessingcanalsobeconductedonthesameimage.2.2BriefreviewaboutJPEGcompressionanddecompressionPriortopresentingourmethod,webrieflyreviewthestandardJEPGcompressioninthefollowingfoursteps.1ARGBcolorimageisfirstlyconvertedtoYCbCrformat,containingoneluminancecomponentYandtwochrominancecomponentsCbandCr1.2Theimageisthensplitintononoverlappingblocksof8x8pixels.Ineachblock,theYCb,andCrcomponentsundergoaDCTtransformafterallthesepixelsvaluesbeingconvertedfromunsignedtosignedintegerse.g.,from0,255to一128,127.f3EachDCTcoefficientcisdividedbyaquantizationstepqandthenroundedtoitsnearestinteger.41ThequantizedDCTcoefficientsforall8x8blocksareentropyencoded,using,say,Huffmancode.Instep3,alargerquantizationstepgwillyieldaseverercompressionrate,whichmeanslowerqualityofJPEGimage,andviceversa.Becausehumanvisualsystemismuchmoresensitivetosmal1variationsoverlowfrequencycomponentsthantothatoverhighfrequencycomponents,themagnitudesofthehi曲frequencycomponentsarestoredwithaloweraccuracy也anthatofthelowfrequencycomponents.ThedecodingofaJPEGimageinvolvesstraightforwardlytheinversestepsofJPEGcompressionentropydecoding,inverseDCTfIDCTandinverseJPEGquantization.Inourmethod,todetectthecopypasteandthetamperedregion.wefirstresavedtheto.bedetectedJ.Cent.SouthUniv.2012l928392851imageinJPEGformatwithdifferentJPEGqualityfactors,andthencomputetheASADimagesbetweentheto_bedetectedimageandeachresavedJPEGcompressedimage.Finally,wejudgewhethertheimagehasbeentamperedand,ifso,determinethetamperedregionbyobservingthegeneratedASADimages.IfanyregionintheASADimagesturnsouttobebrighterordarkerthantherestregionsofthesameASADimage,theregmncorrespondingtotheimageunderexaminationisconsideredtobetampered.Notethatinthefirststepofourmethod,afterbeingresavedinJPEGformat,thetO_bedetectedimageundergoesanotherJPEGcompression.28412.3TamperingdetectionforScenario1Inthissection,weintroducethesimulationcurvesofcopypastetamperingdetectionforScenarioI。andprovideanexampletoillustratehowtodetectthetamperedregion.Tofacilitateouranalysis,weobtainasequenceofcoefficientsthepixelvaluesfroma512512uncompressedgrayscaleLenaimagecolumnbycolumnfrom,e.g.,lefttorightanddenotethesecoefficientsbyCoWeapplydoubleJPEGcompressiontothesecoefficientswiththefirstquantizationstep,g1z,toobtaincoefficientsc1f,andthenthesecondquantizationstep,q2f,toyieldcoefficientsc2f.WecomputetheASADvaluebetweenc2iandc1assADf512512厶一nlc2,z一c1I,f0,1Letgldenoteintegersfrom9to23inanincrementof7,q2aq2Kl6,andq2a一g2g.ThesoliddashcurveineachgraphofFig.1showstheASADvaluebetweenc2andc】c】/7asafunctionofqrangingfrom1to30inanincrementof1.Itcanbeobservedthatcurves6candineachgraphdonotoverlapexceptwheng1n1,asshowninFig.1b.ThecurveCreachesitsminimumifqlngIisintegermultiplesofg,andcurvereachesitsloca1maximumwhenqisdoubleof1d,i.e.q2q218,asshowninFig.1a.FromFigs.1aandfc1。wealsoobserveanothercharacteristicthat,thedifferencebetweencurvesand,denotedbyd,isobviousatsomequantizationstepsq.Itisobservedfromtheabovesimulationthatifthesame1.Ddatasequenceisquantizedwithtwodifferentquantizationsteps,onecandoublyquanfizethesetwosequenceswithaseriesofdifferentquantizationstepsandobtaintwodifferentASADcurves.ThedifferencebetweenthesetwoASADcurveswillberatherobviousatsomequantizationstepsusedinthedoublequantization.Thisdifferenceindicatesthatthe凸矗尝∽昙Quantizationstep,qQuantizationstep,qQuantizationstep,qFig.1ASADbetweendatasequencequantizedbyqxdg1,followedquantizedbyqinrangeof1,30Thedifferencebetweencurvesandisdenotedbyag19bgIn16cql23quantizationstepsusedinthefirstquantizationfortwodatasequencesaredifferent.ThisobservationcanbeusedforthecopypasteforgerydetectionofscenarioI.Figure2showsthetamperedimage,inwhichacentralportion,havingbeenJPEGcompressedatquality兰二tamperedsincethegrayvaluesofthecentralregionandtheremainingregionaredistinctlydifferentintheASADimage.28432.4TamperingdetectionforScenarioIISimilartoScenarioI,weprocessdatasequence,Co,usingthefollowingprocedurefirstlytheyarequantizedbyq】,subsequentlyquantizedbyg2toyieldcoefficientsc2,andlastlyquantizedbyq3itoyieldcoefficientsc3i.TheASADvaluebetweenc3fandc2fiscomputedas1512x512VASAO∽,TherearetwodifferentcasesinScenarioII乱sminOIA,Olgandb.Qs_minQ1A,Q1BWeexplaintheprocessbyillustratinganexample.Letqlgl11andq123.Letq2aq2flbeintegersfrom7to15inanincrementof4,andq3q3gbeintegersfrom1to30inanincrementof1.InFig.3,thesoliddashedcurve。【showstheASADvaluebetweenc3c3pandc2c3∥asafunctionofq.ItisobservedthatbothASADvaluesineachgraphreachtheirminimaVASAD0whengIisintegermultiplesofq.Mostimportantly,thedifferencebetweenCHIVES仅andisobviousatsomequantizationstepsg,suchasfrom20to30inFigs.3aandb,andfrom25to30inFig.3c.Thedifferencebetweencurvesdandreachesitsmaximumatg。二23asshowninFig.3b,whichismainlybecausethequantizationstepql/J23isnearlydoubledasthatofq2p11.Figure4ashowsthetamperedimage,whichwasproducedbyinsertingaregionfromanimagewithJPEGqualityfactorofQ1A85intothecentralportionofanotherimagewithJPEGqualityfactorofQ1B55,andsavingthecompositeimageatJPEGqualityfactorofQs75.ShownineachsubsequentpanelistheASADimagebetweenthetamperedimageandaresavedJPEGcompressedimageatdifferentqualityfactorsQrschangingfrom30to60inanincrementof5.Figure4clearlyindicatestheevidenceofcopypastetamperingconductedbetweenJPEGimages.Caseb.QminQ1A,Q,BSimilartoCasea,letql11,g1口23,g2g2changefrom25to29inanincrementof2,andq3aq3pqchangefrom1to30inanincrementof1.WeobservefromFig.5that,ineachgraph,curvesandalmostchangesynchronouslyandoverlap,andtherefore,thedifferencebetweencurvesandissmallforal1ofthequantizationstepsq.Therefore,itseemsprohibitedtodetectthetamperedregioninthiscasebytheproposedmethod.曼凸矗Quantizationstep,qQuantizationstep,qQuantizationstep,qFig.3ASADsimulationcurveswithq2q2psmallerthanmaximumofqlandqlflaq27bq211Cq2152.5ProposedmethodAccordingtothesimulations,analysesandexamplesabove,weproposethecopypasteforgerydetectionmethodconsistingofthefollowingsteps.1JPEGcompression.WeresavethetobedetectedimagefusingJPEGqualityfactorofQrs,toobtainaJPEGcompressedimage尼.1∞J.Cent.SouthUniv.2012192839285128452∽暑∽尝Quantizationstep,qQuantizationstep,qQuantizationstep,qFig.5ASADsimulationcurveswithq2glargerthanqldandgl口aq225bq227cq22931Deterrninewhethertheimagehasexperiencedcopypastetamperingornotand,ifso,determinethetamperedregionbyobservingthegeneratedASADimage.IfanyregionintheASADimagestumsouttobebrighterordarkerthantherestregionsofthesameASADimage,theregioncorrespondingtotheimageunderexaminationisconsideredtampered.f4、RepeattheabovestepsforanotherdifferentresavedJPEGqualityfactorQrsifneeded.Inourmethod,thevalueofblocksizebcanbearbitrarilychosen.Ifbistoosmall,thedetectedtamperedregionwillnotbeclearlyenoughtodifferentiateitfromtherestregion.Ifbistoolarge.theedgeofdetectedtamperedregionwillbeblurrythesmallertamperedregionswillbedifficulttodetectandthecomputingtimewillincreaseremarkably.Inthispaper,bischosenas10.Andwewilldiscussthesizeofbindetailsintheexperiments.3ExperimentalresultsanddiscussionWestartwithoriginaluncompressedimages,eachofsize576x768,fromRe251.EachofthetamperedimagesisobtainedfromtwoJPEGimagesofdifferentqualityfactorsandsavedinanuncompressedformatfortamperingScenarioIorintheJPEGformatfortamperingScenarioII.Figure6ashowsanoriginalimagewithJPEGqualityfactor60.Figure6bshowsatamperedimagegeneratedbyinsertingatravelerfromaJPEGimagewithqualityfactor80,andbycoveringanothertravelerwhowasabitfarawayfromthecamerawithasmallpatchofsceneryfromaJPEGimagewithqualityfactor80,andthensavingthecompositeimageasaJPEGimagewithqualityfactor90.EachsubsequentpanelistheASADimagebetweenthetamperedimageandaresavedJPEGcompressedimageatdifferentqualityfactorsfrom52to64inanincrementof4.ItisclearfromtheseASADimagesthatthegivenimagehasbeentamperedandbothtamperedregionshavebeendetectedeveniftheyaresmallinsize.Figure7ashowsatamperedimageobtainedbyinsertingashrubfromaJPEGimageofqualityfactor10/12AdobePhotoshopCS4intotheoriginalimageFig.7bofJPEGqualityfactor4/12andsavingthecompositeimageasaJPEGimagewithqualityfactor8/12.EachsubsequentpanelistheASADimagebetweenthetamperedimageandaresavedJPEGcompressedimageatdifferentqualityfactorsfrom60to90inanincrementof10.Figure8showsatamperedimageproducedbyinsertingacalffromaJPEGimagewithqualityfactor10/12intotheoriginalJPEGimageofqualityfactor6/12,andsavingthecompositeimageasaBMPformat.ThelasttwopanelsinthefirstrowofFig.8fromlefttorightshowstheBAGimageandthemarkedBAGimage,respectively,generatedbyusingthemethodinRe6.Theirresultsshowthetamperedregionsaswellassomenontamperedregions.EachrowofthebottomportionsaretheASADimagesbetweenthetamperedimageanda

注意事项

本文(Passive detection of copy-paste forgery between JPEG images)为本站会员(dingyx0101)主动上传,人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。 若此文所含内容侵犯了您的版权或隐私,请立即通知人人文库网([email protected]),我们立即给予删除!

温馨提示:如果因为网速或其他原因下载失败请重新下载,重复下载不扣分。

copyright@ 2015-2017 人人文库网网站版权所有
苏ICP备12009002号-5