国内图书分类号：TP309.7国际图书分类号：621.3硕士学位论文基于生物特征的数字水印技术研究硕士研究生：贺夏龙导师：曾国坤申请学位：工程硕士学科、专业：计算机技术所在单位：深圳研究生院答辩日期：2013年12月授予学位单位：哈尔滨工业大学万方数据
ClassifiedIndex:TP309.7U.D.C:621.3DissertationfortheMaster’sDegreeinEngineeringRESEARCHOFDIGITALWATERMARKINGWITHBIOLOGICALFEATURESCandidate:XialongHeSupervisor:KuoKunTsengAcademicDegreeAppliedfor:Master’sinEngineeringSpecialty:ComputerTechnologyAffiliation:ShenzhenGraduateSchoolDateofDefence:December,2013Degree-Conferring-Institution:HarbinInstituteofTechnology万方数据
摘要随着科技的推广和进展，人们对于生物特征信息的认识越来越全面。很多生物特征被人们用作进行生物识别。因此，在如今开放的网络中，对于生物特征信息的保护也变得特别的重要。本课题主要研究了能够应用在生物特征领域的数字水印技术。除此之外，还涉及到数据压缩，加密算法和特征识别等一些技术。在经过对国内外的生物特征水印以及相关心电压缩识别等研究状况的学习后，本文主要做了如下四个方面的研究工作。（1）首先是关于心电图特征信息方面的数字水印。主要研究了一种带自同步信号的心电特征水印算法，它是基于小波分解的。该方法在嵌入同步代码的同时，把水印序列嵌入到心电平滑区的低频小波域中，从而提高了信息隐藏的鲁棒性和同步代码的搜索效率。（2）该部分主要研究了在数字水印的前提下，心电压缩和识别的算法。当然这些算法也是基于小波的。该研究改进了一种高压缩率的有损压缩方式，同时，在使用小波方法进行识别时，通过改进取窗口函数的方法，提高了识别的效果。（3）该部分研究了一种使用磁性写板收集信息的多人图像数字水印技术。通过获取人们的手写特征信息，并对其进行提取。在嵌入方面主要采用了最低有效位算法，然后增加了对特征信息的幻方加密，从而在特征上对它进行了改进。（4）在课题最后的部分，提出了一种综合心电水印和多人手写数据的图像水印架构，其中很多算法是基于上述步骤。在效果评估方面，本文选取了一些有代表性的算法作为对比。这些算法被用来代替本文的算法进行测试。通过对比不同方法得到的参数和结果，研究中采用了诸如信噪比，误码率，压噪比，峰值信噪比等等一些参数。同时，还设计了一些其他方法的对比，在控制变量的前提下，对不同方法进行比较。另外，还对数据结果进行模拟攻击。在这些情况下，我们的算法大部分时候都表现出了较好的性能。关键字：手写签名；生物水印；心电水印；文档水印万方数据
ABSTRACTWiththeprogressanddevelopmentofscienceandtechnology,people’sknowledgeofthebiometricinformationbecomesmoreandmoremature.Anincreasingnumberofbiologicalcharacteristicsareusedtocarryoutbiologicalrecognition.Thistopicmainlystudiesthedigitalwatermarkingtechnologythatcanbeusedinthefieldofbiologicalcharacteristics.Inaddition,italsoinvolvesthedatacompression,encryptionalgorithmsandsomefeaturerecognitiontechnologies.Aftersurveyingtherelatedresearchesaboutbiologicalcharacteristicsofwatermark,ECGcompressionandrecognition,thispapermainlydealtwiththefollowingfouraspects.(1)ThefirstoneisdigitalwatermarkingonECGbiometricinformation.ItmainlymakesaresearchonanECGcharacteristicswatermarkingalgorithmwithself-synchronoussignalwhichisbasedonthewaveletdecomposition.ThismethodembedsthewatermarkingsequencesinthesmoothareaofECGatthesametime.Inthisway,therobustnessoftheinformationhidingandsearchefficiencywiththesynchronizationcodehasbeenimproved.(2)ThesecondtopicmainlyfocusesonECGcompressionandrecognitionalgorithmwiththepremiseofdigitalwatermarking.Ofcourse,thesealgorithmsarebasedonthewaveletaswell.Theresearchimprovedalossycompressionmethodwithhighcompressionratio.Meanwhile,itimprovesthewindowfunctionmethodtoenhancetherecognitioneffect.(3)Thethirdresearchproposesadigitalimagewatermarkingtechnologybasedonmulti-peoplehandwritingsignatures.Itfirstpickuppeople'shandwritinginformationandextractedimportanteigenvalues,andthisembeddingalgorithmisbasedontheleastsignificantbitmethod.Moreover,itenhancesencryptionabilitywiththemagicsquareencryptionoffeatureinformation.(4)Inthefinalofthetopic,itproposesanewwatermarkarchitecturewhichcombinedECGwatermarkandhandwritingsignature.Itisanintegrationoftheabovechapters.Atthesametime,theresearchhasalsohascarriedontheexperimentwiththenewapproach.Inthecomprehensiveassessment,thisthesisselectssomerepresentativemethodsascomparisonalgorithms.Throughthecomparisonontheparametersandresultsofdifferentmethods,thisthesisobtainsmoreconvincingresults.Somecriteriaareusedintheevaluation万方数据
suchasSNR,BER,andCNR,PSNRandsoon.Inaddition,thesimulatedattacksonthesedata.Forcomprehensiveevaluationastheresults,ouralgorithmshowsbetterperformance.Keywords:handwritingsignature,biologicalwatermarking,ECGwatermarking,documentwatermarking万方数据
ACKNOWLEDGEMENTSAfternearlyoneyearoflearningandresearch,Ifinallycompletedthegraduationdesign.Inthewholeresearchprocess,Iencounteredmanydifficulties.Butwiththehelpoftheteachersandclassmates,Ifinallyfinishedthem.Iobtainedalotfromthisgraduationdesign,includingtheabilityofthinkingandworking.Intheprocess,mysupervisor,teacherKuo-KunTseng,caredmuchaboutmygraduationdesign,althoughheisverybusy.Dr.Tsengprovidesusthegoodenvironment,butalsogivingusprofessionalguidance.Here,IwanttoexpressmysincerethankstoDr.Tseng.Atthesametime,myclassmateswhoareinthesamegroupwithme,ChenHeandLiYifan,havegivenmegreathelp.Iamgratefulforthem.Atthesametime,IalsowanttoexpressmythankstoalltheteachersandstudentswhoareleadedbyProf.PanZhengxianginlaboratoryIIIRC.Theyareallgivingmeagreatdealofencouragementinmyworkandstudy.万方数据
CONTENTSPageListofTables..................................................................................................................XListofFigures...............................................................................................................XI1.Introduction.............................................................................................................11.1Thebackgroundandmotivation...............................................................11.1.1ECGwatermarking,compressionandidentification.....................11.1.2Multiplayersdocumentswatermark...............................................21.2Researchcontents.....................................................................................31.2.1ECGwatermarking........................................................................31.2.2ECGcompressionandidentification.............................................31.2.3Multiplayersdocumentswatermarkandnewschema...................41.3Structureoftheorganization....................................................................42.AnewwaveletbasedwatermarkingforECGsignals..........................................62.1RelatedworkaboutECGwatermarking...................................................62.2Proposedarchitectureandalgorithm........................................................72.2.1DataPreparation............................................................................72.2.2WatermarkingEmbedding.............................................................92.2.3Discretewavelettransforms.........................................................102.3Evaluation...............................................................................................132.3.1Performance.................................................................................132.3.2Comparedwithothermethods.....................................................152.4Summary.................................................................................................223.CompressionforECGsignalswithverificationevaluationafterwatermarking233.1Relatedwork...........................................................................................233.2Proposedarchitectureandalgorithm......................................................243.2.1Wavelettransformsofdatacompression.....................................25万方数据
3.2.2Waveletidentification..................................................................273.3Evaluation...............................................................................................273.3.1PerformanceEvaluation...............................................................283.3.2Comparisonwithothermethods..................................................333.3.3Verificationevaluationwithimprovedwaveletmethod..............363.4Summary.................................................................................................374.Multiplayer-documentswatermarksignaturewithhandwritingfeatures......384.1Therelatedworkaboutimagewatermarking.........................................384.2Backgroundofdigitalwatermark...........................................................394.3Proposedarchitectureandalgorithm......................................................414.3.1Maintainingtheintegrityofthespecifications.............................414.3.2ImprovedLSBalgorithm..............................................................434.3.3Themagicmatrix..........................................................................464.4Evaluation...............................................................................................474.5Summary.................................................................................................515.Twostagesdocumentwatermarkingwithfrequencybasedalgorithm............525.1Introductionandapplication...................................................................525.2Thenewarchitecture..............................................................................535.2.1Background...................................................................................535.2.2Themodeloftwostageswatermarking........................................545.3Summary.................................................................................................566.Conclusion..............................................................................................................57References......................................................................................................................59Appendices....................................................................................................................62万方数据
LISTOFTABLESTablePageTable2.1SNRcomparedbeforeandaftermodification........................................13Table2.2ComparewithDCTSNR........................................................................16Table2.3Watermarkperformances.......................................................................16Table2.4Robustnessofwatermark.......................................................................17Table2.5ComparisonofSNR...............................................................................19Table2.6Robustnessoftheirmethod....................................................................19Table2.748setsresults..........................................................................................21Table3.1Parametercomparisonwithdifferentorder............................................29Table3.2Parametercomparisonwithdifferentfunction.......................................29Table3.3Compressionratio...................................................................................33Table4.1ComparisonofdifferentmethodsPSNRvalue......................................50Table4.2PSNRcomparisonofdifferentdata........................................................50Table4.3ThePSNRvalueunderthecapacitytest................................................50Table4.4ExtremedocumentimagetestPSNR......................................................51万方数据
LISTOFFIGURESFigurePageFigure2.1RelatedworksaboutECGwatermarking.................................................6Figure2.2ECGschematicdiagram...........................................................................8Figure2.3Embeddingwatermarkmodel...................................................................9Figure2.4Detectingwatermarkmodel......................................................................9Figure2.5Decompositionlevel...............................................................................11Figure2.6Embeddingmodel...................................................................................11Figure2.7Extractingmodel....................................................................................12Figure2.8TheembeddedSNR................................................................................14Figure2.9Watermarkedandsourcesignal..............................................................14Figure2.10Mergerwatermarkedandsourcesignal..................................................15Figure2.11BERcomparisonsoftwomethods.........................................................20Figure2.12Thetrendchart........................................................................................21Figure3.1Relatedworksaboutcompression..........................................................23Figure3.2Compressionmodel................................................................................26Figure3.3ThenoiseSNR........................................................................................30Figure3.4ThedecompressionSNR........................................................................30Figure3.5Thecompressionnoiseratio...................................................................31Figure3.6TheoriginalandwatermarkedECG.......................................................32Figure3.7ThedecompressionECG........................................................................32Figure3.8OurDWTmixedfigure..........................................................................34Figure3.9FFTmixedfigure....................................................................................34Figure3.10DCTmixedfigure...................................................................................34Figure3.11Thebiterrorrate.....................................................................................35Figure3.12Recognitionsuccessrate.........................................................................36Figure3.13Datacomparisonchart............................................................................37万方数据
Figure4.1Watermarkmodel...................................................................................39Figure4.2Documentauthenticationsystemarchitecture........................................42Figure4.3Theprocessofwatermark.......................................................................43Figure4.4Testoriginalimage.................................................................................48Figure4.5LSB,DCT,DWTwatermarkedimage...................................................49Figure4.8Testfullgridgraph.................................................................................51Figure5.1Hospitalapplicationdiagram..................................................................53Figure5.2Relatedworksaboutthemodel..............................................................54Figure5.3Thenewmodeloftwostageswatermarking..........................................55万方数据
Chapter1IntroductionCHAPTER1INTRODUCTION1.1ThebackgroundandmotivationWiththerapiddevelopmentofinformationandcomputernetworks,digitalmedia[1]exchangeofinformationhasbecomeanessentialpartofmodernlife.Thedigitalmediaisnotonlybringingusconveniencebutalsocarryingrisks.Individuals’privacyisattractingmoreandmoreattention.Electrocardiogramsandotherbiologicalfeaturesarebeingapplied[2]moreandmoreasabiometricpersonaldataanddeservetobeprotected.Atthesametime,theuseofthedigitalmediaisincreasinganditscarryingcapacityisbeingtestedunprecedentedly.Forexample,inthepast,thepatientdatawasrandomlystoredinthe[3]hospitalwithoutanyprotection.However,withthedevelopmentofscienceandtechnology,itwasfoundthatsomepatientdatacontainsimportantprivateinformation,andsomeofthe[4]informationcanevenbeusedbyvillains.Therefore,protectionmeasuresshouldbetakenintheprocessofstoring,transmitting,orbrowsingtheinformation.Otherwise,manyunnecessarytroubleswillariseonceinformationleakageorinformationattackoccurs.Thedigitalwatermarkingtechnologywhichisbasedonthebiologicalcharacteristicshasrapidlydevelopedinthisenvironment.Thistopicisproposedandstudiedinthiscontext.1.1.1ECGwatermarking,compressionandidentificationAnECGreflectstheprocessoftheelectricalactivityofheart,whichcanbetakenasa[5]referenceforthestudyofcardiacfunctionandcardiacpathology.WithanECG,wecananalyzeandidentifyvariousarrhythmias,andunderstandthedegreeanddevelopmentofmyocardialdamage,aswellasthestructureandfunctionoftheatriumandventricle.Thisresearchuseselectrocardiography(ECG)data.AnECGmaybeusednotonlytoanalyzedisease,butalsotoprovidecrucialbiometricinformationforidentificationandauthentication.Therefore,thesubjectproposesamethodbasedonwaveletsinthisarticle,toaddwatermarks1万方数据
Chapter1Introductiontotheelectrocardiogramandcompressthem.Thus,thesubjectexpectstosavetheECGcharacteristicspremisewhileprotectingthesecurityoftheECGdatainnetworktransmission.Thesubjectperformedthefrequencydomaintothestratifieddata,andthenembeddedthehiddeninformationbitinthelowfrequencysub-bandofECGsmoothareawithDWTdomain,sothatitcanwithstandattacksandinterference.Aftertheinversewavelettransform,weagainperformedastratificationofthefrequencydomain,whilethehigh-frequencycomponentsettledtozero.Aftertwocompressions,wetransmitthedata,anddoaninversetransformtoobtainthewatermarkeddata.Inordertosimulatetherealenvironment,wedesignthemethodtoaddingnoiseattacks.Theexperimentalresultsshowthatthehiddendatahasbetterrobustness.1.1.2MultiplayersdocumentswatermarkThedocumentinformationcopyrightprotectionproblemsarebecomingincreasinglyserious.Itisacrossinterdisciplinarycontainscryptography,interdisciplinary,visualsciences,graphics,Chineseinformationprocessing,imageprocessing,communications,and[6]InformationSecurity.Intheprocessofaddingwatermark,itwillembedsectionofflag[7]copyrightinformationintotheprotectedmediainformation.Onlytheowneroftheintellectualpropertyrightscandeterminewhetherthedigitalwatermarkexists.Althoughinrecentyears,digitalwatermarkinghasachievedagreatdevelopmentinthetheoryand[8]application,butithasn’tformedacompletetheoreticalsystem.Inparticular,thereisnouniformcriterion;therearestillmanyunresolvedissues.Therefore,thedigitalwatermarkingtechnologyisavibrantareaofresearch.Inordertosolvetheproblemofinformationsecurityandcopyrightprotection,digitalwatermarkingtechnologycameintobeing.Withtheongoingresearchofdigitalwatermarkingtechnology,peoplefoundthattosolvethecopyrightissuesofdigitalproductsanditsdifferentstagesoftherelease,sale,use,andsooncopyrightauthenticationissues,[9]peopleneedmorewatermarkinformationembeddedtothesamedigitalproducts.Multiplayerdigitalwatermarkingtechnologyisusetosolvetheproblemofcopyrightissues[10]aswellasmanypeopletosignthedocumentcopyrightauthentication.Thereforemultiplayerdigitalwatermarkindigitaldocumentcopyrightprotectionandownershipidentificationhasgoodpracticalvalue.2万方数据
Chapter1Introduction1.2Researchcontents1.2.1ECGwatermarkingThissubjectpresentstheECGbasedonorthogonalwaveletdomainwatermarkingencryptionalgorithm.Weusethesynchronizationcodeinthehiddeninformationbitstomakethehiddendataachieveself-synchronization.SynchronizationcodeandhiddeninformationbitsareembeddedintotheDWTdomainofthelow-frequencysub-band,makingitabletowithstandtheattackandinterference.Weusethetime-frequencylocationcharacterofwavelettransform,reducingtheburdenofsearchingsynchronizationcode.Theexperimentalresultsshowbetterrobustnessofhiddendata.Inthelast,thissubjecthastheSNR(signalnoiseratio)andBER(biterrorrate)analysisoftheprogram,evaluationoftheoveralleffect.Forthis,itusesanevaluationformulatoconnecttheSNRandembeddeddatatoensurethetransparencyoftheembeddeddata.BERinGaussiannoisecanbeusedtoevaluatetheperformanceofthedesign.Sincetheexperiments,theembeddeddatahassufficientrobustnesstowithstandtheattackandinterference.1.2.2ECGcompressionandidentificationThissubjectproposesanewelectrocardiogramcompressionmethodafterthewatermark.ThesubjectusesawaveletcompressionmethodtohandletheECGsignalafterthewatermark.Inthepremisewithoutchangingthewaveform,thecompressionrateismaintainedat25%ofthestate.Byremovingthehighfrequencyportion,wecanensurethecompressionrateandaccuracy.Aftercompression,itcanbealignedfortransmission.Afterthat,weapplytheBER,CNR(thecompression-to-noiseratio)andseveraldifferentperiodsofSNRtoassessthealgorithm,andthencomparetheresultwithotheralgorithms.Thefinalresultsprovethatouralgorithmisrobustandfeasibleinthecomprehensiveevaluation.Finally,weuseaverificationmethodthatwedesignedtoidentifythesignalidentificationbeforeandafterwatermark.Inverificationmethodsection,thesubjectalsoputsforwardtheinnovativepractice,especiallyintheareaofthewindowfunction.3万方数据
Chapter1Introduction1.2.3MultiplayersdocumentswatermarkandnewschemaThissubjectgothandwritteninformationandembeddedthemasthewatermarkintothedocumentpicture.Inthisstep,itusesalosslessembeddingmethod.Intheend,thewatermarksignalwasextractedtobeidentifiedandauthenticated.Throughouttheprocess,themainmethodisLSB(LeastSignificantBit)methodforthewatermark.Inordertoprotectthedocumentsandthesecurityofthehandwritteninformation,wedesignedthisscheme.Atthesametime,severalcomparativealgorithmsexperimentsaredesignedtodetecttheprogramresults.Notonlythat,inordertoimprovethesafetyperformanceofthedata,thestudyusedthemagicsquaretransformationtoencryptinformation.Thenthesubjectselectedevaluationcriteriawhichnowarepopulartoevaluatetheeffects.Forexample,filtering,pepperattacks,PSNR(PeakSignalNoiseRatio),etc.Aftertheexperiment,theresearchproposedanewwatermarkingarchitecturebasedonthesetwosteps.Thenewarchitectureusestwolevelswatermark.First,itaddedawatermarktotheECGsignalasEigenvalues.Then,itembeddedtheECGsignalasthewatermarkinformationintothedocument.Inthesecondstep,thearchitectureusedalosslessembeddingmethod.Intheend,thewatermarksignalwasextractedtobeidentifiedandauthenticated.1.3StructureoftheorganizationThefirstchapterintroducesthemaintothewhole,theresearchbackgroundandresearchcontent.ThesecondchaptermainlyexplainedtheECGwatermarkingalgorithmbasedonwavelettransform.TheresearchembeddedwatermarkIDtosmoothECGarea,soastoreducetheimpactonECGwave.ThischapteralsoincludestheintroductionofrelatedworkaboutECGwatermark,algorithminterpretationanddataanalysis.ThethirdchaptermainlyrecordsresearchesaboutECGdatacompressionandrecognitionalgorithm.Ofcourse,thesewereconductedonthebasisofECGdigitalwatermarking.ThestudyimprovedDWTcompressionmethod,andoptimizedtheDWTrecognitionmethod.Intheaspectoffindingwindowfunction,bystretchandcuttingtheECG,theresearchimprovedtheefficiencyofrecognition.Thefourthchapterintroducedtheresearchaboutpeopledocumentwatermarkingtechnologybasedonthecharacteristicsofhandwritten.ThisresearchembeddedwatermarkinginformationintotheimagebymeansofnewLSB.Amongthem,anewextractionmethodwas4万方数据
Chapter1Introductiondesignedinaspectofdataacquisitionandabstractedthehandwrittencharacter.Atthesametime,usingofmagicsquaretoencryptdataimprovedthereliabilityofthedata.Thefifthchaptercombinedthepreviousresearchesandproposedanewmethodofarchitecture.AfterthefirststepbyECGwatermark,thenthemethodgotitandwritefeaturesfordatafusion,andaddedthemtotheimagewatermark.Themethodputforwardthedoublewatermarkstructurebasedonbiologicalcharacteristics.Thisisanewauthenticationmodel.Thesixthchaptermainlyconcludedthepreviouswork,andintroducedtheunfinishedworkandsubsequentwork.5万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsCHAPTER2ANEWWAVELETBASEDWATERMARKINGFORECGSIGNALS2.1RelatedworkaboutECGwatermarkingAtpresent,fromthewatermarkaspect,researchontheprotectionofECGinformationisstillinitsinfancy,althoughtherearesomeresearches,showninfigure2.1,relatedtothewatermarkingofECGsignals,andwiththeuseofwavelettransformbaseddigitalwatermarkingencryptiontechnology.Therefore,researchinthisfieldhasgreatpotentialfortheresearcher.Theexistingresearchesmaybedividedintothefollowingcategories:SecureMedicalImageAnalysisSecureDataWatermarkECGWirelessAlgorithmsTransmissionCompressionSecureECGItselfIdentificationFigure2.1RelatedworksaboutECGwatermarkingThefirstapplicationisthedigitalwatermarktechnologyusedinmedicalimages.Thisapplicationproposesanovelblindwatermarkingmethod,namelyembeddingasecretkeyinto[11]themedicalimageofECGsignals.ThesecondisasensornetworkbasedECGmonitoringsystem.ECGsignalsarewatermarkedwithpatientbiomedicalinformationtoconfirmpatient/ECGlinkageintegrity.Itisinfinitudeofsignaldistortion,whichissufficienttohold[12]thepatientinformationwithoutaffectingtheoverallqualityoftheECG.ThethirdapplicationiswavelettransformbasedECGdigitalwatermarkingtechnology.Thismainly6万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsappliestoECGsignals.InECGsignals,theenergyisconcentratedinQRScomplexwaves.SotheselectionofwaveletcoefficientsforconcealmentshouldavoidcausingtheQRS[13]complexwavestodistortobviously.ThelastapplicationisECGtransmissioninawirelessnetwork.ThisoneproposestheuseofdigitalwatermarkingtoensurethesafetransmissionofECGsignalsinawirelessnetwork.Alowfrequencychirpsignalisusedtoembedthe[14]watermark,whichisthe15-bitsdigitalcodeofthepatient.Thecharacteristicoftheproposedwatermarkingschemeisthattheembeddedwatermarkcanbefullyremovedbythereceiverduetotheblindrecoveryfeatureofthewatermark.2.2ProposedarchitectureandalgorithmThemethodsmentionedinthisarticleoriginatedinthepaper“EfficientlySelf-SynchronizedAudioWatermarkingforAssuredAudioDataTransmission”,writtenbyShaoquanWu,JiwuHuang,DarenHuang,Shi,Y.Q.Theirmethodisusedforaudio[8]encryption,toensurereliableaudiotransmission.Afterlearningtheirideasandmethods,weachievetheirmethodandthenimprovedit.ThismakesthismethodmoreinlinewiththerequirementsoftheECG.ThefirstthingtodoisreadingtheECGsignal.Andthisisdifferentforreadingtheaudiosignal.Nowadays,therearethreeinternationallyrecognizedECGdatabasescanbeused.TheirnamesaretheMassachusettsInstituteofMIT-BIHelectricaldatabase,theAHAdatabaseoftheAmericanHeartAssociationandtheEuropeanST-TECGdatabase.WewillusetheMITdatabase.Ineachdata,weselectedthelengthofthe4096partofthedataforanalysis.Wemodifythesamplingrateto360.AndthenwehavetheprocesstoeliminateDCoffset,andstandardize.Finallywesettimeindex,thelengthoftheECGdataandthesizeoftheembeddedwatermark.Beforetheextractionofthewatermarkweaddedastep.WeaddedwhitenoiseattacktotheECGsignalofthewatermark.2.2.1DataPreparationECGreferstotheheartineachcardiaccycle,whichthepacemaker,atrialandventricularsuccessivelyexcitedonebyone,alongwiththebioelectricalchangesinECG,ECGtracingsleadstothegraphicsofthevariousformsofpotentialchangesfromthesurface[15](referredtoasECG).ECGistheheartexcitedabouttheoccurrence,spreadandrecovery7万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsprocessobjectiveindicators.ECGisexcitedheartelectricalactivity,andithasanimportantreferencevalueinbasicfunctionsoftheheartandpathologyresearch.ECGcanbeusedtoanalyzeandidentifyavarietyofarrhythmias;itcanalsoreflecttheextentanddevelopmentof[16]myocardialdamageandatrialandventricularfunctionandstructuralcondition.Ithasthereferencevalueinguidingcardiacsurgeryandinstructingnecessarydrugtreatment.Figure2.2ECGschematicdiagramThestandardECGleadtoelectrocardiogramwaves,namedbytheDutchphysiologistW.Einthoven(theinventoroftheECGWilliamEinthoven).HedividedonecardiaccycleintoP,Q,R,S,Twaves.Pwave:theexcitementoftheheartoriginatesinthesinusnode,andthenreachestheatrium.Pwaveisproducedbytheatrialdepolarization.Itisthefirstwaveofeachwave[17]group.Itreflectsthedepolarizationprocessoftheleftandrightatrium.ThefirsthalfofthePwaverepresentativestherightatrium,thelatterpartofthePwaverepresentativestheleftatrium.QRScomplex:atypicalQRScomplexincludesthreecloselylinkedwaves.ThefirstdownwardwaveiscalledQ-wave,followingahigh-tip-Q-waveverticalwaveknownastheRwave.ThedownwardwavebyRwaveiscalledS-wave.Becausetheyarecloselylinked,andreflecttheexcitementoftheventricularelectricalprocess,itiscollectivelyreferredtoastheQRScomplex.Thiswavegroupreflectstheleftandrightventriculardepolarization[17]process.Twave:TwaveislocatedinfollowedSTsegment.Itisarelativelylowand[18]accountedformuchlongerwave.Itisgeneratedbyventricularrepolarization.8万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsAccordingtotheabovedescription,theECGdiagnosismainlydependsonthePQRSTwaves.Therefore,whenweaddwatermark,tomaintaintheshapeofthesewaveformsisverynecessary.2.2.2WatermarkingEmbeddingDigitalwatermarkingtechnologyreferstodirectlyembeddingsomeidentifyinginformation(digitalwatermark)intothedigitalcarrier(includingmultimedia,documents,software,etc.),whichdoesnotaffecttheusagevalueoftheoriginalcarrierandishardtobe[19]perceivedornoticedbypeople'sperceptionsystem(suchasvisualorauditorysystem).Theinformationhiddeninthecarriercanhelpusconfirmthecontentcreators,buyersandcarrierstransmitsecretinformation,anddeterminewhetherthecarrierisaltered.Digitalwatermarkingisanimportantresearchdirectionofinformationhidingtechnology.ECGhashighrequirementsforaccuracyandthewatermarkedimagecanbehighlyrestoredwithoutbeingdamaged,sowatermarkingencryptionisareasonableselection.Herearethefigure2.3and2.4ofembeddinganddetectingwatermark:Figure2.3EmbeddingwatermarkmodelFigure2.4Detectingwatermarkmodel9万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsInthisprocess,itusedthesynchronizationcode.Theprincipleisasshown2.6.Itcanbeusedtolocatehiddeninformation,topreventtheunpredictableattacks.Theformulaisasfollows2.1.AssumethatP1isapositiveerrorrate,P2isanegativeerrorrate,listhelengthofthesynchronizationcode,eisthedeterminedthreshold.∑(2.1)∑(2.2)2.2.3DiscretewavelettransformsThisresearchmainlyusedwavelettransformtohandleasegmentationofECGsignal.2ThewavelettransformmapsafunctioninL(R)ontoascale-spaceplane.Waveletsareobtainedbyasingleprototypefunction(motherwavelet)ψ(x)whichisregulatedwitha[20]scalingparameterandashiftparameter.Thediscretenormalizedscalingandwaveletbasisfunctionaredefinedas2.3and2.4.(2.3)zhikuquan20150807(2.4)Whereiandnarethedilationandtranslationparameters;hiandgiarethelow-passandhigh-passfilters.Orthogonalwaveletbasisfunctionsnotonlyprovidesimplecalculation2incoefficientsexpansionbutalsospanL(R)insignalprocessing.Asaresult,anydigital2signalS(t)∈L(R)canbeexpressedasaseriesexpansionoforthogonalscalingfunctionsandwavelets.Morespecifically,∑∑∑(2.5)Where∫(2.6)And∫(2.7)Theydenotethesequencesoflow-passandhigh-passcoefficients,respectively;j0istheintegertodefineanintervalonwhichS(t)ispiecewiseconstant.Throughoutthispaper,thehostdigitalECGsignalS(n),n∈R,denotingsamplesoftheoriginalECGsignalS(t)atthenthsampletime,iscutintosegmentswhereDWTwillbeperformed.ThiscanbedonebyexploitingtheHaarwaveletwithanorthogonalbasistoimplementDWTthroughafilterbank.10万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsFigure2.5demonstrateshowtheinputdigitalECGsignalS(n)isdecomposedintoeightnon-overlappingmulti-resolutionsub-bandsbytheseven-levelDWTdecomposition.Figure2.5DecompositionlevelFigure2.5illustratestheprocessthatusing7levelswaveletdecompositiontodividetheinputECGsignalsintoeightnon-overlappingsub-bands.Takingintoaccounttherobustperformanceofthelow-passfiltering,weembeddedsynchronizationcodeandwatermarkintotheseventhlevelofthesub-bandcoefficient,thatthelowestfrequencywaveletcoefficients.Aftermixingupsynchronouscodewithpartofembeddedinformation,embedlowfrequencysubbandDWTfactorofprocessedECGaswatermarking,thenobtainzhikuquan20150807watermarkedECGthoughinversetransform.AsshowninFig2.6showstheembeddingmodel.Figure2.6EmbeddingmodelIndataextractionpart,firstneedtoexecutethesameDWTwiththatusedin[21]embeddingwatermark,andextractbinarydatafromlowfrequencysubbandDWTfactor.Thenobtainsynchronouscodefromextracteddata.Whileselectingforsignalmovably,theprocedureneedstoconstantlyrepeatexecution,untilsynchronouscodeisdetected.Thedetailedprocessisdescribedasfollows.Afterensuringthelocationofsynchronouscode,afterdeterminethelocationofthesynchronizationcode,wecanextractthehiddeninformationofthefollowingsynchronizationcode.Extractingmodelshowninfigure2.7.11万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsFigure2.7ExtractingmodelInthissession,wehavetheECGsignalsegmentsandimplementtheDWToneachfragment.Thenthesequence{mi}isembeddedineachlow-frequencysub-band.Signalsegmentlengthdependsonthelevelofwaveletdecomposition.Itshouldbeabletoaccommodateatleastonesynchronizationcodeandsomeoftheinformationdata.Theruleofembeddedasfollows.⌊⌋{(2.8)⌊⌋’Amongthem,ciandciaretheDWTcoefficients,Sistheembeddingstrength.zhikuquan20150807Inadditiontothesynchronizationcode,wealsohavethedatapayloadthatreferstothenumberofbitswhichareembeddedintotheaudio,measuredintheunitofbps(bitper-second)andusedintheformulaasB.AssumethatthesamplerateisR(Hz),WaveletdecompositionlevelisK.Theformulainthealgorithmasfollows.(2.9)Whenextractingthedata,wedividetheECGsignalwhichhavebeenembeddedwatermarkintoseveralfragments,ofwhichatleastincludesonesynchronizationcode[22]segment.Thenweperformedwavelettransformoneachsectionwiththesameas*embeddingwatermark.Suppose{ci}isthecoefficientoflow-frequencysub-band,weusethe**followingrulesextractsequence{mi}from{ci}.⌊⌋{(2.10)⌊⌋12万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignals2.3EvaluationInthissection,theresearchhasanormalECGdatawhichwerecollectedfromtheMIT-BIHArrhytmiadatabase,andusingourmethod,weexecutewatermarkencryptionontheseECGdata.Sothat,wecanevaluatetheembeddeddatatransmissionrateofsynchronizationcodeandthedatapayloadofthisalgorithm.WemainlyusetheSNRandBERtocarryouttheevaluation.Inwhichtheywillbeproposedintheinformula2.11and2.12.Acharacteristicofthewatermarkencryptionmethodistransparenttousers.Thepresenceofthewatermarkdoesnotaffecttheusers.Therefore,weshouldmaintaintheconsistencyofthewatermarkedsignalandtheoriginalsignalinthemaximumextentpossible.2.3.1PerformanceFirst,letuslearnthedefinitionoftheSNRandBER.∑∑(2.11)(2.12)’Amongthem,the{fi}and{fi}representsthesourcesignalandthemodifiedsignal.WeselectedfoursetsofdatafromtheMIT-BIHdatabase.Thenwedoouralgorithmonthesesignals.First,weusetheimprovedmethodtocomparewiththepreviousmethod.Wesetseveraldetectingmultiplicationfactorsasthemagnificationofwhitenoise.Theyare1,50,500and1000.HereistheSNRcomparisonbeforeandaftermodificationofmethods.Asisshowninthetable2.1below:Table2.1SNRcomparedbeforeandaftermodificationDataBeforeSNRAfterSNRClass127.238230.4968Class225.638230.6291Class329.774331.7761Class428.362932.0753Itcanbeseenfromthetable2.1,afterwemodifiedtheoriginalmethod,theSNRvalueisdefinitelyimproved.Thisshowsthatafterthemodified,thismethodismoresuitablefortheECGdataofthewatermark,wecangetbetterresults.13万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsWeselectedeighteensetsofdatafromtheMIT-BIHdatabase.Thenweappliedouralgorithmonthesesignals.First,weusedtheimprovedmethodtocompareitwiththepreviousmethod.Wesetseveraldetectingmultiplicationfactorsasthemagnificationofwhitenoise.Theseare10,50,450and950.Figure2.8showstheSNRofthesignalsafterwatermark.Thehorizontalaxisshowsdatasets,longitudinalaxisshowsSNRvalues.SNRvalues4030SN20R100123456789101112131415161718DatasetsFigure2.8TheembeddedSNRAscanbeseenfromthechartabove,theSNRvalueafteraddingthewatermarkisdistributedatabout30.Inthethreemethodswehaveused,theportionofthewatermarkisthesame.Accordingtothestateofknowledge,theSNRvalueisintherangeofabout30.Wecanassumethattheresultshavebeenverygood.Thechangeswillnothavemuchimpactonthewaveformandwillnotcauseerrorsinthedoctor'sdiagnosis.Asthefigures2.9and2.10shownfollows,bluecurverepresentsthesourcesignal;greencurverepresentsthewatermarkedsignal.WatermarkedECGOriginalECG110.80.80.60.60.40.4Amplitude0.2Amplitude0.200-0.2-0.2-0.4-0.412345678910111234567891011Time(sec)Time(sec)Figure2.9Watermarkedandsourcesignal14万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignals1.2OriginalECGWatermarkedECG10.80.6Amplitude0.40.20-0.20.10.20.30.40.50.60.70.80.91Time(sec)Figure2.10MergerwatermarkedandsourcesignalThatcanbeseen,accordingtotheimage,thedifferencebetweenthewatermarkedsignalandthesourcesignalisverysmall,almostnothing.TheECGsamplingratethatweusedis360.Ineachdata,weselectedafragmentlengthof4096tobetested.Synchronizationcodeweusealengthof63msequenceand256-bitwatermarksequence.Wedefinedthresholdas21.TheHaarwavelettransformthatweusedhaseightdecompositionlevels,asmentionedin2.2.3.2.3.2ComparedwithothermethodsWehavechosentheDWTasthemainpartofthealgorithm.Thisisveryconducivetotherealizationofthesynchronizationcode.Inourexperiments,thesynchronizationcodeisusedtolocatethepositionofthewatermark.Timedomainandfrequencydomainisthebasicnatureofthesignal,sothatyoucanuseavarietyofwaystoanalyzethesignal,andeach[23]providesadifferentperspective.Thetimedomainandfrequencydomaincanbeawareoftheinteractionbetweentheresponsesignalsandinterconnect.Thetimedomainisthereal[24]world,itisonlytheactualexistenceofthedomain.Ifwewritesynchronizationcodesequenceinthetimedomain,andthewatermarkisembeddedinDCTcoefficients,bytime-domainembeddingstrengthlimit,therobustnessofthesynchronizationcodewillbelimitedtoo.TotheFFT,thewaveletfunctionisdiversity;FFTfunctionisrelativelysimple.On[25]specificissues,DWTismoresuitableforanalysis.Frequencydomainisnottrue,itisamathematicalconstruct.Ontheotherhand,ifthesynchronizationcodeisembeddedintothe15万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsfrequencydomain,suchasDCT,DFTdomain,comparedtoDWTdomain,theircomputationalcostoflookingupthesynchronizationcodeswillbegreatlyincreased.AccordingtotheprincipleoftheDCT,weaddawatermarktotheECGdatausingtheDCTmethodexperiment.TheSNRweobtainedisshownintable2.2below.Table2.2ComparewithDCTSNROurSNRDCTSNRClass130.4968-3.4942Class230.6291-2.2896Class331.77610.2802Class432.0753-2.9436Ascanbeseenfromthetable2.2,SNRvalueisseriouslylessthantheexpectedresult,obtainedbytheDCTmethod.BecausewhenweuseaDCTtransform,theECGimagebecomesoffset.ThisdirectresultsthedecreaseoftheSNRvalue.Table2.3WatermarkperformancesDataFsTimeSNRlevelClass13601030.49687Class23601030.62917Class33601031.77617Class43601032.07537Thetable2.3aboveisatestofthefoursetsofdataandweobtaineddifferentSNRvalues.Inthissetting,itcanbeseenthattheSNRvaluesabove30shouldbeabletomeettheaccuracyrequirements,thatdoesn’timpactthedoctors’diagnoseofECG.Afterthat,wetestthewatermarkedsignalwiththewhitenoiseattacks.Forthis,wecanunderstandtherobustnessofwatermark.Inaddingawatermarkingsignal,weartificiallyaddedwhitenoise,tosimulatetransmissionchannelnoiseandthepossibleattack.Specificdataareasfollowstable2.4.16万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsTable2.4RobustnessofwatermarkDataNoiseBERErrorSNR_noiseClass110072.6337Class1500038.6543Class15000018.6543Class1100021.875712.6337Class210071.8875Class2500037.9081Class25000017.9081Class2100021.875711.8875Class310073.9786Class3500039.9992Class35000019.9992Class3100021.875713.9786Class410073.7127Class4500031.3995Class45000019.7333Class4100021.875713.7127Ascanbeseenfromthetable2.4,thecommonnoisedidnoteffectonthewatermark.Thesedata,whenthewhitenoisefigureis1,50,500,BERis0.Whenwhitenoisefigureis1.50,theSNRwithnoiseisabove30.Intoday'swidespreaduseofADSLbroadbandnetwork,itgenerallydoesnotproducenoise.Whenthenoiseabout5dB.Whenthenoiseupto30-50dB,thenetworkhasbeenbasicallyunusable.Whenthenoiseisabout5dB,thenetworkhasalreadybegunwithanunstablecondition.Atthispointthenetworkwillbeassessedasthereisaseriousnetworkfailure.Onlywhenthenoiseisbigenough,thatismorethan500times,itwillappearsomedeviations.However,suchabignoiseinrealitydoesnotseemore.Therefore,therobustnessofthewatermarkisgood.Inthenextpart,wein-depthstudytheotherarticlesontheECGwatermark,andselecttherepresentativemethodtocompare.Wefocusonlearningtheideasandmethodsofthesearticles;weachievethemalongourunderstandingofthem.Accordingtotheparameterslisted17万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsinthearticle,wetrytotestthedatatomaintainaconsistentlevelwiththeoriginalexperiments.Andthismakestherealizationthatwehavedonebenottoofarawaytheoriginalauthorsreferredto.Aftertheexperiment,wecanseethatourexperimentaldataobtainedandtheoriginaldataandresultstheauthorslistedarealsosimilar.Althoughtheremaybesomedeviation,itisinevitable.However,inthefollowingexperimentalresults,Iwouldliketobesufficienttointroducewhatweexpect.Theoneis“WaveletTransformationBasedWatermarkingTechniqueforHumanElectrocardiogram(ECG)”;theauthorsareMehmetEngin,OğuzÇıdamandErkanZekiEngin.Thecentralideaofthisarticleissimilarwithourmethod;thereisthevalueofthecontrast.First,theyusethediscretewavelettransformtodecomposeanECGsignalintoeightsub-bands.Andthencalculatetheaveragepowerofeachsub-band.Atlast,theyinsertedtherandomsequencethatisthewatermarksignal,intotheselectedsub-bandofthem.Inthisprocess,theyusedwaveletfunctionDaubechies(db2)andBiorthogonal(bior5.5).Inthewatermarkembeddingprocess,theyusearandomlygeneratedGaussiannoisefigure,andaTheravadacoefficientasthecoefficienttogeneratearandomsequence.Theembeddedprocessissimplythecumulativeprocess.Inthewatermarkextractionprocess,theyneedtheoriginalsignalasaninputtosolvethewatermarksequence.Webelievethatthiswilllosethesignificanceofnetworksecurity,andinthisway,wecannotguaranteethesecurityoftransmission.IfthesourceECGsignalandthewatermarkECGsignalaremodifiedatthesametime,thentheremaynotbedetected.Inthedatarespect,theyselectedfourtypicalsetofdata,includinganormalelectrocardiogramanddiseaseelectrocardiogram.TheirdataalsocomesfromtheMIT-BIHArrhytmiadatabase.Afterlearningthismethod,wecarriedoutitandimplementedit.Intheoriginalarticle,theycontrastthemethodsdb2andbior5.5.Theyhaveproventhatbior5.5isworsethandb2,soweonlytriedthedb2.Themethodweusecomparedwiththeirs,mostnotablywithself-synchronizationcapabilities.Moreover,intheprocessofwatermarkextraction,wedonotrequireaccesstosourceECG.Inthetable2.5below,withthemethodweimplementedofthemandours,wetestedthesamedata,andthenwefirstgivecomparisonoftheSNRintwomethods,aftertheprocessofembeddedwatermark.18万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsTable2.5ComparisonofSNRDataOurSNRTheirSNRClass130.496820.91Class230.629120.3561Class331.776121.265Class432.075324.98Itcanbeseenfromthetable2.5,withthesamedata,theSNRvalueofourapproachissignificantlyhigher.Thisshowsthattheaftertheprocessofembeddingwatermark,thenoisegeneratedbyourmethodhavefewer.Table2.6RobustnessoftheirmethodDataWhiteNoiseBERErrorSNR_noiseClass1128.125972.6197Class15031.251038.6403Class150037.51218.6403Class11000501612.6197Class2156.251871.9227Class25053.1251737.9433Class250053.1251717.9433Class2100071.8752311.9227Class3162.52073.9056Class35062.52039.9262Class350056.251819.9262Class3100053.1251713.9056Class4156.251873.6919Class45056.251839.7125Class450065.6252119.7125Class4100065.6252113.6919Asthetable2.6shows,itissimilartoourexperiments,weusetheirmethod,addthewhitenoisetotheembeddedwatermark.AndthengivesanSNRvalueasareference.Afterextractingthewatermark,weaddtheBERtoassesstherobustnessofthemethod.According19万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalstothedataweobtained,wecomparedthemwiththedataintheprevioustable2.4.Ascanbeseenfromthedata,whenweaddthesamenoisecoefficient,theerrorratewhichusesourmethodtoextractthewatermarksequenceisrelativelylow.Asshownbelowinfigure2.11,thisisalinechart.ItvisuallyshowstheBERcomparisonoftwomethods.Theerrorrateofwatermarksequenceusingourmethodisrelativelylow.TheirBEROurBER80.0070.0060.0050.00BER40.0030.0020.0010.000.001111222233334444DatasetFigure2.11BERcomparisonsoftwomethodsWedotheuniformtestofall48setsdataintheMIT-BIHArrhytmiadatabaseatthelastpartoftheexperiment.Usingourmethod,wegotthedatasetasshowninthetable2.7below.SNRisthesignaltonoiseratiowhenwehavenotwatermarkedtheECG.WhileSNR_noiseisthedatasignalnoiseratioafterweincreasedthewatermarkandnoiseattack.Here,noiseattackinmultiplesvalueis500.20万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignalsTable2.748setsresultsSNRSNR_noiseSNRSNR_noiseSNRSNR_noise10030.496818.654311734.564823.073821233.204422.727610129.015717.908111836.123524.198621333.907623.017910229.719.999211931.578220.879621434.418923.051210331.399519.733312134.967521.799921534.324523.21610430.992720.039612233.932922.689321738.180425.883710532.693621.873712328.872418.287721933.749921.780610633.524621.047812430.546819.975522030.673819.428410737.564626.296620032.944622.409222132.277821.458710832.53422.83120133.210120.163922230.361419.871410934.768124.933520232.285919.893722335.076921.453611133.327221.117220334.953223.948522833.400223.062811233.480522.071820529.959819.385823031.262620.742411331.587321.005420736.276225.033823131.62719.929211431.988221.625620836.348124.546123232.255420.325911530.24918.494220931.011420.421723334.955224.731611633.684521.160721034.970422.363523429.867220.7317Immediatelyfigure2.12belowisthetrendchartofthedatainthetable2.7above.Thehorizontalaxisshowsdatasets,longitudinalaxisshowsSNRvalues.SNRSNR_noise4540353025SNR20151050100102104106108111113115117119122124201203207209212214217220222228231233DatasetsFigure2.12Thetrendchart21万方数据
Chapter2ANewWaveletBasedWatermarkingForECGSignals2.4SummaryInthisresearch,weproposeaself-simultaneousECGdigitalwatermarktechnologybasedondiscretewavelettransformtoensurethesecurityoftransmissionorpreservationofECG.Inordertoenhancetherobustness,weembedthesynchronizationcodesequenceandthewatermarksequenceintothelowestfrequencyofthewaveletcoefficients.Inordertoimprovetheefficiencyofsynchronizationcodesearching,weusethewavelettransformtime-frequencypositioning.IntheexperimentalresultswecanseethatthewatermarkedECGsignalhasahighSNRvalues,thatareresistanttocommonattacks,andthereisnotmuchchangebythewaveform.Itwillnotaffectdoctorsdiagnosingtheillness.Intheexperiment,weselectedafewrepresentativesECGsignal,hopingtogetageneralconclusion.However,comparedtothelargeECGdatabase,thisisjustoneofasmallpart.Atthesametime,Ithinkthattheexperimentalresultscanalsocontinuetostrengthenthroughthedifferentwaystogetbetterresults,butIalsohopethatthistechniquecanbeappliedtoothercontent,suchasbrainwaves,andthesemayalsobemyfollow-upresearch.22万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingCHAPTER3COMPRESSIONFORECGSIGNALSWITHVERIFICATIONEVALUATIONAFTERWATERMARKING3.1RelatedworkTherearecurrentlynoworksonECGwhichincluderesearchonbothwatermarksandcompression.However,therearesomestudieslookingatcompressionorwatermarkingindividually.Sincethewatermarkinghasbeensurveyedinthelastchapter,inthisresearch,wesurveyedtheotheraspectoncompressionasthekeypoint.Therelatedworksareshownasfigure3.1.Figure3.1RelatedworksaboutcompressionIntheaspectofECGcompression,theECGisadynamicsignal.Itwillcontinuetoproducenewsignals.Forexample,Holtermonitoringtechnologyhasbeenappliedmoreandmore,andtherearepatientsforwhommorethan24hoursofECGdatahastobecollected,whichgreatlyincreasestheamountofdatayouneedtorecord.Withtheadventoftheaging23万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingsociety,thenumberofpatientswithheartdiseasewillgrow,andcardiaccarewillbecomeasocialproblem.RemotetransmissionofECGscanallowreal-timemonitoring;itisconducivetodiagnosisandfirstaidinstructions.Therefore,theremotetransmissionofECGshasagoodeconomicandmarketoutlook.ECGsignalcompressionisakeytechnologyforremoteECGtransmission.Itdirectlydeterminesthepracticalityandeffectivenessofthesystem.Forexample,inawirelesscommunicationnetworkwhichisaccountingfordatatransmission,longtermECGguardianshipgeneratesahugeamountofdatathatwillmakewirelesscommunicationcostsunacceptable,andraiseissuesoftransmissionspeedandbandwidth.ECGsignalcompressiontechnologywillguaranteethatnoneoftheinformationoftheECGsignalislostandwillminimizetheamountofdatathatneedstobetransmitted,reducetransmissioncosts,andincreasetransmissionspeed.Withtheinterventionofcomputertechnology,ECGdatacompressiontechnologyisincreasinglyshowingitsimportance.TheHolterdatacompressionalgorithmisoneofthehotspotsofthecurrentinternationalresearchinthefieldofbiomedicalsignalprocessing.Itisfruitful.Datacompressionispossiblewithavarietyofmethods.Earlypredictivecodingmethods,suchasDifferentialPulseCodeModulation(DPCM),directlyencodetheamplitude[26]variationoftheadjacentsamplevalues.Theprincipleofthesemethodsissimpleandeasytoimplement,butthecompressionrateisrelativelylow.Run-lengthcoding(RLC)usesthecorrelationamongthesymbols,byrecordingthelengthofeachsymboltoachievecompression.Shannon-FanocodesandHuffmancodesarebasedonthefrequencywithwhich[27]eachsignalappears.Thentheyassignthemosteconomicalcodelengthsoastoachievecompression.Withaflatdistributionofthesignalinthetimedomain,afterorthogonaltransformation,theenergywillbeconcentratedonthelow-frequencycomponentsothehigh-[28]frequencycanbeomitted.Orweuseonlyafewbitstoencodethem.ThesetransformcompressionmethodsincludetheKarhunen-Loevetransform(KLT),FourierTransform(DFT,[29]FFT),anddiscretecosinetransform(DCT).Newcompressiontechniquesincludetheneuralnetworkandwavelettransform(DWT)methodsandothers.3.2ProposedarchitectureandalgorithmThefirstthingtodoistoreadtheECGsignal.TherearecurrentlythreeinternationallyrecognizedECGdatabaseswhichcanbeusedasastandard,namely,theMassachusetts24万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingInstitute’sMIT-BIHArrhythmiaDatabase,theAHAdatabaseoftheAmericanHeart[30]Association(AHA)andtheEuropeanST-TECGdatabase.WewillusetheMITdatabase.Foreachdatapacket,weselectedalengthof4096bitsofdataforanalysisandmodifiedthesamplingrateto360.ThenweranaprocesstoeliminateDCoffset,andtostandardize.Finallywesetatimeindex,thelengthoftheECGdataandthesizeoftheembeddedwatermark.Thenwecarriedouttwowaveletcompressions.AfterfinishingeachDWTconversion,wesetthecontentsofthehigh-frequencycomponenttozero.Inthiswayweobtainedanabatementofhalfoftheamountofdata.Beforeextractionofthewatermark,weaddedawhitenoiseattacktotheECGsignalofthewatermark.3.2.1WavelettransformsofdatacompressionSignalprocessinghasbecomeanimportantpartoftheworkofcontemporaryscienceandtechnology.Theaimsofsignalprocessingare:accurateanalysis,diagnosis,compressioncodingandquantization,storage,andrecoverysignal.Currently,theidealtoolforstationarysignalanalysisisstilltheFouriertransform.However,inpracticalapplications,thevastmajorityofsignalsarenon-stationary,soFourieranalysisisnotsuitable.TheECGsignalmentionedinthisarticleisatypicalnon-stationarysignal,andwavelettheoryofmulti-resolutionanalysisforECGsignalprocessingisanewidea.Comparedwithothertime-frequencyanalysis,theadvantagesofwavelettheorycannotonlyadapttothetime-frequencyresolutioncharacteristicsofnon-stationarysignalsbutalsodecomposesignalsonanorthogonalbasis.Anditiseasywithasmallnumberofparameterstodescribethenon-stationarysignals’time-frequencycharacteristics,allconstituteexcellentfeaturesforanextractionalgorithm.Throughoutthisprocess,oncewedocompress,thedatawillbereducedbyhalf.Aftertwicecompressions,thedatacompressionratiois25%.Bycomparisoninthenextsection,wecanseethatusingourmethod,itcannotonlyimprovethedatacompressionratio,butalsobesuperiorineffect.Todothecompression,weselectaprocessingandanalysismethodbasedonwavelettransformoftheECGsignalcompression,andacompressionalgorithmbasedonthebuildingofthesignal,usingnumericalsimulationMATLABsoftware.Thefollowingfigureisacompressionflowdiagramofthesignalthatwehavecreated.Ingeneral,throughtothehigh-frequencyprocess,wecancarryoutdatatransmission.25万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingFigure3.2CompressionmodelInthefirstcompression,weusewaveletfunctionsbior1.1.Inthesecondcompression,weusewaveletfunctionbior3.7.Therein,biorcanalsobeexpressedasbiorNr.Nd.NrandNdarerelatedparametersforremodellinganddecompositionfilterlength.Thebiorwaveletisabiorthogonalwavelet.c1(n)(n∈Z)istheinputofthefilter,i.e.,awatermarksignal.Aftertheconversion,themiddleoutputis∑̃(3.1)∑̃(3.2)Basedontheoutputofthefiltergroup̃∑(3.3)Withmergerandexchange,wecangetthefollowingformulã∑∑̃̃(3.4)Inordertofullyreconstruct,evenwiththẽ(3.5)weneedthefollowingequationtobeestablished.∑[̃̃](3.6)Thewaveletmethodforcompressionplaysaroleinoptimizingthewaveformontheelectrocardiogram.Wechoosetoremovethehighfrequencycomponents.Thechangeinthegeneratedwaveformhasnoeffectonthedoctor'sdiagnosis.Instead,itispossibletomakethewaveformmoreeasilyidentifiable.Inthisway,wereducenotonlythetransmissionvolumebutalsothenoiseoftheECG.26万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarking3.2.2WaveletidentificationThisresearchwatermarkeddatabeforeandafterthe18setsofauthenticationcomparison.Threedifferentmethodsatthesametimethedataisverified.Theverificationprocesscanbedividedintothefollowingthreesteps:Firstofall,wemuststreamlinematchbinarydata.AllECGsignalsarenotstationary.AnECGmayincreasethevalueonthetimeaxis,mayalsobereduced.WecutandsegmentedECGsignal,convertittoabinarysignaloflengthconsistent.Here,weuseasamplingrateof360everyfourcyclesasagroup,everyfourpacketsformarelativelysequence.Ifthecycleisinsufficient,thecopyofthefirstcycleuntilthereissufficientdata.However,theECGsignaleachcycleisnotconsistent,soherewehavetobestretched.Inshortthedata,firstselectthemeanposition,andthedatapadding.Thenlookforalargeraverage,andthenfilled.Untilalllengthsofthecyclesareconsistentwiththelongestone.Secondstep,wehavethesegmentedsignalwavelettransform.Ineachroundoftransformation,wetakeitslow-frequencypart,anddothenextlevelwavelettransformwithit.Until8layerstransformationiscomplete.Atthispoint,we'vegotdatathatwewilldealwith.Thefinalstep,wewanttoprocessthedata.AssumingtwoECGsignalE1andE2.Theyareindifferentsections.Wehavedesignedamethodtodeterminetheirsimilarity.HereisourweighteddistanceformulausedtodeterminethesimilaritysignalE1andE2.∑(3.7)∑Wherein,ErepresentsanECGsegmentofonegroup,whichisthebasicunitfordataanalysis.Srepresentstherelativefrequency,andRrepresentstherankofiinthesequenceE.iis1or2.3.3EvaluationInthissection,wetakenormalECGdatacollectedfromtheMIT-BIHdatabase,andusingthemethod,weexecutewatermarkencryptionandcompressionontheseECGdatapackets.WemainlyusetheSNRandBERtocarryouttheevaluation.BERhasbeenproposedinthepreviousone,asmentionedinformula2.12.Acharacteristicofthewatermark27万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingencryptionmethodisitstransparencytousers,whoarenotaffectedbythepresenceofthewatermark.Therefore,weshouldmaintaintheconsistencyofthewatermarkedsignalandoftheoriginalsignaltothemaximumextentpossible.Compressionhasadoubleeffect.Thefirstistoreducethesizeofthefiletransfer.ThesecondistoenhancethefeaturesoftheECGdata.Atthesametime,wehavealsointroducedanewevaluationstandardCNR(CompressionNoiseRatio).3.3.1PerformanceEvaluationInordertomeasurethequalityofthewatermarkedECGsignal,thesignaltonoiseratio(SNR)isadoptedrespectively.ThedefinitionofSNRhasbeenintroducedinthelastchapter.DuetothefactthattheDWTcoefficientsareimplementedwithorthogonalwaveletbasesandaccordingtoParseval’stheorem,theenergyinasignalisgivenasfollows.∫∑∑∑(3.8)Thoseofthehighfrequencysub-bandarenotaffectedbytheproposedembeddingalgorithmandonlythelowestDWTcoefficientsareadjusted,andhence∑∑()(3.9)∑||∑∑WeapplythisformulatoadjustaproperembeddingstrengthT.Inaddition,evaluationstandardCNR(Compressionnoiseratio)isdefinedas3.9follows.∑(3.10)∑Wheresrepresentsthesourcedataand,Drepresentsthedataaftercompression.Accordingtoourexperiments,whenthewatermark,weselected7-layerwaveletdecompositionforthewatermark,weandobtainedthebestresults.Thisisbecauseweusedthebiorwaveletsfunction.Generally,thefilterlengthis8,andtheordernumberis7.Inthetablebelow,weusethesamesetofdata.Finally,weaddthesamenoisefigureas450.Wejustmodifytheorderandthenobtainedasetofdata.Ascanbeseenfromthetable,ina28万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingdifferentorder,althoughtheSNRandthecompressionnoiseratiochangesaresmall,itshowedadownwardtrend.Withthedecreasingorder,theerrorrateinthefinalBERsurges.Table3.1ParametercomparisonwithdifferentorderSNR_embedCNR1CNR2SNR_noiseSNR_comBERlevel30.496816.256314.717957.018414.84010727.323616.063614.576357.000614.71669623.722515.65814.288757.003314.42698521.058315.162813.954857.056114.039964Aftertheabovesteps,wedeterminedthedifferentwaveletcompressionfunctions.Asshowninthefollowingtable,weselectedfourdifferentwaveletfunctionsforthecompressionoperation.ThetableliststhevaluesoftheCNR1,CNR2andSNRfordifferentsituations.Wherein,SNR_embedindicateswatermarkednoiseratio.CNR1representafterthefirstcompressionmeans,wehaveobtainedacompression-noiseratio.CNR2secondcompressedrepresentation,wehaveobtainedcompressionnoiseratio.SNR_noiserepresentsnoiseratiooftheaddingnoise.SNR_comrepresentsthenoiseratioaftertwicecompression.BERrepresentsthebiterrorrateofdataafterbeingrestored.Atthispointweobtainedcompressionratiois25%.Ascanbeseenfromthedatainthetable3.2,afterthecompressionofthesecondarydecomposition,thebiormethodweusedshowedbetterperformance.Table3.2ParametercomparisonwithdifferentfunctionCNR1CNR2SNR_noiseSNR_comBERFunction16.256314.717957.018414.84010bior16.25639.210756.67959.67190haar16.25639.210756.67959.67190db123.181613.502356.981513.6610coif1Thedatainthefollowingchartfigure3.3istheSNRvaluesobtainedbythedifferentmethodswhenincreasingdifferentnoisevalues.Therein,AistheSNRpolylinewhenthenoisecoefficientis10,BistheSNRpolylinewhenthenoisecoefficientis50,CistheSNRpolylinewhenthenoisecoefficientis450,andDistheSNRpolylinewhenthenoisecoefficientis950.ThegreenlinerepresentstheDWTmethod,theredlineshowstheDCTmethodandthebluelinerepresentstheFFTmethod.29万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingFigure3.3ThenoiseSNRAscanbeseenfromthefourgraphs,thehorizontalaxisshowsdatasets,longitudinalaxisshowsnoisesSNRvalues.Regardlessofhowgreatthenoisecoefficientbe,ourmethodissignificantlybetter.Thefollowingfigure3.4isobtainedfromthedataafterthecompressionrecoveryvalueofthecalculatedSNR.Thehorizontalaxisshowsdatasets,longitudinalaxisshowsdecompressionSNRvalues.ItcanbeseenthattheSNRvaluethatisobtainedbyourmethodwassignificantlyhigherthantheothertwomethods.FFTSNR_comDCTSNR_comDWTSNR_com18161412S10N8R64201234DatasetsFigure3.4ThedecompressionSNR30万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingThefollowinglinegraph3.5showsthecompressionnoiseratio.Thehorizontalaxisshowsdatasets,longitudinalaxisshowsCNRvalues.Sinceourmethodiscompressedtwice,thesetwocompressionnoiseratiosareobtainedseparately.Aftereachonecompression,theamountofdatawillbereducedbyhalf.Aftertwicecompression,thedatacompressionratiois25%.Thegreenlineindicatesthevalueafterthefirstcompressionandthepurplelinerepresentsthevalueafterthesecondcompression.Afterthesecondcompression,theCNRvaluesdecreasedalittle,butwerestillhigherthantheCNRvaluesobtainedbytheothertwomethods.WiththeFFTandDCTmethod,weonlyhaveoncecompression.Inotherwords,thedatacompressionratioisonly50%,obtainedbytheFFTandDCT.TheCNRobtainedbythetwomethodsislowerthanourmethod.Inthiscase,ithasbeenunabletobecompressedagain.OURCNR1OURCNR2FFT_CNRDCT_CNR3025201510CNR50-5123456789101112131415161718-10DatasetsFigure3.5ThecompressionnoiseratioInadifferenttest,weselectedtheimageofasetofdataasshowninthefigures3.6and3.7below.Thebluecurveindicatesthesourceimage,thegreencurverepresentsthewatermarkedimage,andtheredcurverepresentstheimageaftercompression.31万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingFigure3.6TheoriginalandwatermarkedECGFigure3.7ThedecompressionECGAscanbeseenfromtheimages,thedifferencesamongtheoriginalsignal,thewatermarkedsignalandthedecompressionsignalaresosmallastobealmostnegligible.TheECGsamplingratethatweusedwas360.Thecompressionrateis25%.Weusebiorwaveletsfordecompositionandremovethehigh-frequencycomponenttoachievethepurposeofcompression.Intermsofcompressionefficiency,weputitintotwopartstoexperiment.Ifweneedtoconsidertheintegrityofthewatermark,asmentionedabove,thecompressionratiocan32万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingreach25%.So,ifwejustcompressitasrecognitiondata,compressionratiocanreachoneoversixteen.Itisshowninthetable3.3below.Table3.3CompressionratioConsiderwatermarkConsiderverificationCompressionratio25%6.25%Sincethat,atthetimeofrecognition,wejustcutoutthefragmentswithmorecharacteristicvalues,thePQRSTpartineachcycle.Andthen,wewillextendittothesamelength,128bits.Usefourcycles,theamountofdataisonly512bits.Plusthepreviouscompressionmethod,finally,itcanreachoneoversixteen.3.3.2ComparisonwithothermethodsAfterobtainingthedataweresearchedthemethodtogenerate.Westudiedsomeotherclassiccompressionmethodsandransomecomparativetests.Afterin-depthdiscussion,wesettledontheDCTandFFTmethodsandthencontrastedthesewithourmethod.IntheFFTandDCTcompressiontest,weselectedthesameparametersasforDWT.TheformulaforthedefinitionoftheFFTmethodis:∑(3.11)∑(3.12)Where(3.13)TheformulafortheDCTmethodis:[∑∑](3.14)Weusethesamecompressionratiocontrolat50%.ForFFTandDCT,afterconductingthetransformonce,weuseanintervalabatementapproach.Wesettheneareroneofthevaluestozero,soastoachievethepurposeofcompression.Atthereceivingend,wedotheinversetransform.Thedataisrestoredtoitsoriginallength.Thefollowingthreefigures3.8,3.9and3.10contrasttheimagesfromthethreemethods.Thewatermarksectionusesthesamemethod,theHaarwavelettransformfunction.Sotheoriginalsignalandthe33万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingwatermarkedsignalofthreeimagesareconsistent.Theseimagesarecutoutoftheentireimages,displayingperiodrangesinordertomoreeasilyobservethedifferentpoints.Figure3.8OurDWTmixedfigureFigure3.9FFTmixedfigureFigure3.10DCTmixedfigureAscanbeseenfromtheaboveimages,whenusingtheFFTmethod,thedecompressedimagepeakbecomeshalfoftheoriginal.Althoughtheoverallwaveformdoesnotshowtoo34万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingmanychanges,butinsomelocations,itappearsthatthephenomenonofadoublecrestappears.Inthisimage,theECGdataiscompressedby50%.Thisimagecannotbeusedfordiagnosis.WhenusingtheDCTmethod,wecanseethat,justthelocationoftheQRScomplexpeaksarereduced,andthereisnodoublecrestwithouttheoccurrenceofthephenomenonofdoublecrest.However,thepositionoftheTwave,theinherentlylowTwaveisamplified,andevenhigherthantheQRSwavepeak.Thiswillseriouslyaffectthedoctor's[31]diagnosis,resultinginthephenomenonofamiscarriageofjustice.Thefollowingfigure3.11showsthedifferentdatapacketsasindifferentnoisevalues,usingdifferentmethods,fromwhichweobtainedtheresultingbiterrorrate.Thehorizontalaxisshowsdatasetswithdifferentwhitenoise,longitudinalaxisshowsBERvalues.Thesevaluesareobtainedfromtheextractedwatermarksequencesaftercompressionandrestorationcomparedwiththeoriginalwatermarksequences.FFTBERDCTBERDWTBER605040BE30R2010010504509501050450950105045095010504509501111222233334444DatasetsFigure3.11ThebiterrorrateTheerrorratesobtainedusingourmethodaregenerally0.TheBERobtainedusingtheothertwomethodsaregenerallymorethan30%.Withthiserrorrate;thewatermarkeffectisbasicallylost.35万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarking3.3.3VerificationevaluationwithimprovedwaveletmethodInthissection,wehavedesignedanauthenticationmethod.Basedonpreviousexperience,wehavechosenDWTastheauthenticationprocessingcore.Butintheoverallprocess,wemadesomeadjustmentsandchanges,sotheeffectismoreobvious.Thefigure3.12showstherecognitionsuccessrateinadifferentsamplerate,teamandcycle.Thehorizontalaxisshowsdatasetswithdifferentcycles,teamsandsamplerates,longitudinalaxisshowssuccessrate.Thefinalselectionofgroupingis4,theperiodis4,andthesamplingrateis400.Inthiscase,thehighestrecognitionratecanbeobtained,0.9901,where,s-tisthesamplingrate,crepresentsthecycle,tindicatestheteam.Inthismethod,thefirstthingisthedetectionofECGpeaks.AfterfindingPQRSTpeaksvalues,datawillbecutout.Removetherelativelysmooth,lesscharacteristicvaluesareas.Thenstretchthesecyclestothesamelength.Wesetthestandardlengthas128bits.Thesefeaturescanbealsousedoncompression.Inthiscase,ifyoudon'tconsiderthewatermark,compressionefficiencywillbeachievedoneoversixteen.success1.050.990110.950.9Successrate0.850.80.75s-r320360400440480320360400440480320360400440480320360400440t44444888888888810101010c44444444448888810101010SetswithdifferentcyclesteamsandsampleratesFigure3.12RecognitionsuccessrateInthistest,weusedtheeighteensetsofdata.Eachgroupconsistedoftheoriginalsignalandthewatermarkedsignal.Meanwhile,wehavetheDCTmethodandFFTmethodcompared.Hereiswherewegetthedatacomparisonchart.36万方数据
Chapter3CompressionForECGSignalsWithVerificationEvaluationAfterWatermarkingDWTDCTFFT25000200001500010000Similardistance50000123456789101112131415161718DatasetsFigure3.13DatacomparisonchartThefigure3.13showsthreecurvesobtainedbythemethodsimilardistancevalue.Thehorizontalaxisshowsdatasets,longitudinalaxisshowssimilardistancevalues.Whenthisvalueissmaller,itrepresentsmoresimilarandtheidentifiabilityishigher.Ascanbeseenfromthefigure3.13,whenweuseourmethod,theobtainedsimilaritydistanceisminimal.3.4SummaryInthisresearch,wehavedesignedandimplementedawaveletanalysisofanECGwatermarkandcompressionalgorithm.ThistechnologycanbeusedtoprotectECGtransmissionsecurity,andreducethetransmissionvolume,whileoptimizingtheECGshape.Inordertoguaranteethesecurityofthewatermarkembeddedinthesignalwaveletdecomposition,thewatermarkwasembeddedinthelowfrequencycoefficients.Forcompression,wechoseanotherwaveletfunction,andthenweremovedthehigh-frequencypartofthewaveletdecomposition.ItsimpactontheECGshapeissmall.Forexperimentalcomparison,weselectedtheDCTandFFTcompressionmethods.Inthedataevaluation,wechosetheSNR,CNR,andBERasthreeindicatorstomeasure.Inordertotesttherobustnessofthescheme,wealsotestedthesignaltonoiseattack.Weincreaseddifferentnoisecoefficientstosimulatetheactualchannel.Frommanypointsofview,ourmethodstillhassomeadvantages.Nevertheless,thisresearchareaisstillinitsinfancy.Inthefuture,wewillstrivetoimprovetheeffectofthemethodandextenditsapplicationstootherfields.37万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesCHAPTER4MULTIPLAYER-DOCUMENTSWATERMARKSIGNATUREWITHHANDWRITINGFEATURES4.1TherelatedworkaboutimagewatermarkingTheconceptofdigitalwatermarkingisproposedbyCaronniin1993,appliedtothe[32]digitalimage.In1994,VanSchyndelpublishedanarticleentitled"Adigitalwatermark"inICIP'94meeting.Thisisanimportantarticleondigitalwatermarkingpublishedonimportantmeeting.Thearticleclarifiessomeimportantconceptsaboutthewatermark.Itisconsideredaliteratureofhistoricalvalue.Sincethendigitalwatermarkingtechnologyhasbeenpaidmoreandmoreattention.EspeciallysincetheconveningofFirstWorldHideinternationalconferenceintheUniversityofCambridgeduringMay1996,theinformationhiding,especiallydigitalwatermarkingtechnologyhasbeendevelopingrapidly.Digitalwatermark[33]hasimportantprospectsintheeconomic,technicalapplications.Inglobal,manygovernmentagenciesandresearchdepartmentssupportforsuchresearches,includingtheUnitedStatesDepartmentoftheTreasury,theEuropeanTelecommunicationUnion,the[34]MassachusettsInstituteofTechnology,Microsoft,LucentBellLabsandotherinstitutions.Meanwhile,IBM,Hitachi,NEC,PioneerandSonyfivecompaniesalsoannouncedthatthey[35]willjointresearchthedigitalwatermarkbasedoninformationhiding.Atthesametime,somecompanieshavegraduallyintroducedcommercialsoftwaresystemforwatermarkingtechnology.Afterrecentyearsofresearchanddevelopment,digitalwatermarkingtechnologyhasmadeconsiderableprogress.Algorithmfrominitiallysimpleembeddedairspacebasedontheleastsignificantbit(LSB)algorithmhasbecomethemainstreamofthestagebasedonthediscretecosinetransformanddiscretewavelettransformdomainalgorithm.Whilealgorithm[36]theoreticallyprogress,therearesomeproductsandsolutionsbeinglaunched.Itsfieldofapplicationisexpanding.Butingeneral,thedigitalwatermarkingtechnologyasayoungdiscipline,itstheoreticalsystemhasnotbeenperfected;thetechnologyisnotmature[37]enough.Thereisn’tawatermarkyetabletowithstandallknownattacks.Mostalgorithms38万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesarestillexploratory.Theyneedtobetestedinpracticeandcontinuetodevelop.Wehavealongwaytogotoreachawiderangeofapplications.Theseneedresearcherstomakemoreefforts.Sofar,manypapershavedigitalwatermarkembeddingandextractionalgorithm,digitalwatermarkingdigitalwatermarkofthespatialdomainandtransformdomaindigitalwatermarkingcanbedividedintotwomaincategories,dependingonthedigitalwatermarkingmethodofloading.China'sdigitalwatermarkingtechnologyresearchstartedlate,butintheconveningofthefirstsessionsincetheInformationHidingWorkshop,Chinainthefieldofresearchand[38]progressivedevelopment.IncludingTianjinUniversity,BeijingUniversityofPostsandTelecommunications,SunYat-senUniversity,ZhejiangUniversity,Xi'anUniversityofScienceandTechnologyandmanyotherresearchinstitutions,scientistshavecontributedto[39]this.4.2BackgroundofdigitalwatermarkDigitalwatermarkingtechnologyhasbeenintroducedinthechapter2.Butinthischapter,theresearchismainlyabouttheimagewatermark.Thefigure4.1istheflowchartofembeddinganddetectingwatermarkmodelonimagewatermark:Figure4.1Watermarkmodel39万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesThedigitalimagewatermarkembeddedmodelisshowninfigure4.1.Thewatermarkisembeddedintotheoriginalimage.ThesysteminputiswatermarkinformationW,theoriginalvectordataI,andanoptionalK(privatekey/publickey).Whereinthewatermark[40]informationmaybeanyformofdata.KeyKcanbeusedtoenhancesecurity,inordertoavoidunauthorizedrestorationandrehabilitationofthewatermark.Themainconsiderationinthewatermarkembeddingquestionisbywhatmeans,inwhatpositionoftheimagecanbe[41]invisiblyembeddedwatermark,butalsotakeintoaccounttherobustnessofthewatermark.SetErepresentsthefunctionwhichembeddedwatermarkWintotheoriginalimageIusingthekeyK,Iwrepresenttheimageobtainedwithwatermark,thenthewatermarkembeddingcanberepresentedas:(4.1)Theembeddedfunctionscanbeasimplelinearsuperposition.Inanotherway,itcanalsobeavarietyofnon-linearprocessingmethod.Therearethreetopembeddingformulasasthefollowing:(4.2)(4.3)(4.4)Whereinformula4.2istheaddingwatermark;formula4.3isthemultiplicativewatermark;formula4.4istheindexwatermark.Here,W(n)isthewatermarksignalcomponent,αisembeddingstrength.Formula4.2issuitableforthecircumstancesthatI(n)valuechangeslessdramatic.ToasmallerαW(n),theformula4.3and4.4obtainedsimilarresults.Ifthelogarithmoftheformula4.4,itissimilartotheformula4.2.Sotheformula4.4canbeseenasanapplicationoftheformula4.2.Theformula4.3embeddedwatermarkstrengthassociatedwiththeimage.Ithasacertaindegreeofself-adaptabilitycanbeappliedinmostcases.αofoptionsmustbeconsideredforthenatureofimageandcharacteristicsofthevisualsysteminordertoguaranteenotvisibleunderthepremise,asfaraspossibletoimprovethestrengthoftheembeddedwatermark.Inpracticalapplications,theformula4.2and4.3areusedmore.Ascanbeseenfromtheformulaabove,thewatermarkembeddingisusuallyselectedtheembeddedpositionI(n)(pixelorafrequencydomaincoefficientoftheairspace)intheimage.Thenembeddingwatermarkintoimagebasedonthewatermark40万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesfunction.ThenwegettheadjustedvalueIw(n),andI(n)isreplacedbyIw(n)backintotheoriginalimage.NowwecanobtaintheimageIwcontainingthewatermark.Typically,inthemiddleoftwoprocessesofwatermarkembeddingandextraction(ordetected),duringtransmission,thewatermarkedimagemaybesubjectedavarietyofimageprocessingor[42]maliciousattacks.UsingIwdrepresentstheIwaftertheattack.Bothofthemhaveasimilarappearance,orIwdisthedeclineinthequalityofimageIw.Ingeneral,thecommonlyusedmethodforextractingmeaningfulwatermarkobtainedissimilartotheoriginalwatermark,thewatermarkinformation;usuallymeaninglesswatermarkdetectionmethodtodirectlydeterminewhethertheimagecontainsthespecified[43]watermark.Inmanycases,thecorrelationcoefficientbetweenthedetectionresultsbytheoriginalwatermarkandthewatermarkisextractedfromthewatermarkimageindecisionInadditionthestatisticaltestcanalsobeusedinthewatermarkdetection.Letwrepresentsanestimatedwatermark,Dforthewatermarkdetectionalgorithm,Iw'meansnotsubjecttoattackduringthetransmissionorbyawatermarkafterattackvectorimages.Thewatermarkextractionisexpressedas:WithoriginalvectordataI:(4.5)Withoriginalwatermarkw:(4.6)Nooriginalinformation:(4.7)4.3Proposedarchitectureandalgorithm4.3.1MaintainingtheintegrityofthespecificationsThefollowingisthestructureandmethodbasedonhandwritingfeaturesmulti-layereddocumentwatermark.Figure4.2showsthethedocumentwatermarkauthenticationsystemarchitecturebasedonhandwritingcharacteristics.Itisdividedintosixparts:(a)toloadthesourceimage,(b)toloaddata,(c)dataencryption,(d)embedding,(e)thewatermark[44]extraction,(f)datadecryption.Figure3showsthespecificstepsintheprocessaddwatermark.First,theusersignaturetabletwritten,theinformationwillbesavedindividually.Asignatureimagefile,andtheotherisasignatureforeachpointfeatureinformationfile.Then,weneedasigneddocumenttobeconverted.TheformatofthedocumentcanbeDOC,41万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesTXT,PDF,andsoon,afterconversion,wegetaBMPimageformats.Then,weusethethehandwritingfeatureinformationobtaineddocuments,thedocumentembedding.Thedifferentsignaturefeatureinformationloadsignatureinformationhasfinishedloading,youcanloadthesecondperson'ssignature,aperson.Alltheinformationneedtobesignedafterloading,unifiedinformationloadedre-extractedandtheencryptionofthemagicsquare.Andthentheencryptedinformationtothewatermark.Afterthiswecanputthefiletoaneedtoauditdepartmentsorindividuals.Firstimagewatermarkverificationintherecognizedwatermarkunspoiledcase,theycansolvewatermarkprocessingontheimage,thesignaturecharacteristicinformationextracted.Signaturefeatureinformationcanthenrestorepointssignaturestatuscanbeverifiedasahandwrittensignaturerecognition.Thisachievesdoubleprotection.Figure4.2DocumentauthenticationsystemarchitectureAsshowninFigure4.3inthedocumentconvertedtoanimage,wehavetotakesignatureinformationasastreamofdatascrambling.Thenusethatdataasawatermarkdata.Waitforthewatermarkembedding.Eachpointintheacquiredinformationincludesthespecificinformation,includingcoordinates,pressure,andtimeinformation.Thelargestdegreeofreductioncanbecarriedonthehandwrittenhandwrittenownstatebyinformation,42万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesincludingglyphwritingaswellasthethicknessofthestrokes,order,andthespeedofwriting.Werefertothesespecialdatapreserved,thenumberofsignaturesasaKEYisalsoembeddedintotheoriginalimagewhichMixedbyembeddingalgorithm,andfinallygeneratetheimageafteraddingthewatermark.Whentherearemanypeoplesignedthesamedocument,suchasthepartiestothecontractsignatureconfirmation,orsignedbythehigherlevelofthemulti-layerallowscase,wedeterminethemaximumvalueoftheembeddeddatainarange,andaKEYrecordnumberofsignatures.Everyonewilltakeupspaceintheembeddeddata.Figure4.3Theprocessofwatermark4.3.2ImprovedLSBalgorithmAdigitalimageiscomposedofpixels.Eachpixelhasagrayscalevalue;thehigher[45]grayscalevaluesindicategreaterbrightness.Thegrayvaluesaretypicallybetween0and255.0representsblackand255representswhite.Imagegrayvaluescanberepresentedbythe8-bitbinarynumber.Highestbitcontributesthemosttograyvalue;thelowestbitcontributes[46]theleasttograyvalue.Wecalledthelowestbitasleastsignificantbit(LSB).Inthismethod,itextractedthedifferentbitsofallpixelsoftheimage,andthenconstitutedeight43万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesdifferentbit-planes.Wherein,thebitplaneconstructedbythelowestbitsiscalledtheleastsignificantbitplane.Foranaturalimage,thedifferencebetweenadjacentpixelsisoftennotbig,which[47]meansthatthecorrelationofadjacentpixelsisstrong.Thehigherthebitplane,thegreateritcontributestothepixelgrayvalues.Andthus,thecorrelationofadjacentpixelsisstrongerofhighbit,showingabetterregularity.Onthecontrary,thelowerthebitplane,the[48]contributionissmallerofthepixelgray.Therefore,thelowbitoftheadjacentpixelisweak;itissimilartorandomnoise.Theleastsignificantbitplaneoftheimagecontainsthe[49]leastimageinformation,thecorrelationbetweenthepixelscanbeapproximatedasrandom.Thecorrelationofthemostsignificantbitplaneofimageisthestrongest;itcontainsthemostenergyoftheimage.ThebasicwayofLSBwatermarkisusingthesecretinformationwhichwewanttoembedintoreplaceleastsignificantbitofthecarrierimage.LSBwatermarkisasimplebuteffectivewatermarkalgorithm.Thealgorithmrequiresonlychangingverysmallanddifficulttodetectofthecarrierimage,itcanembedinalargenumberofsecretinformationwithhighspeed.LSBwatermarkembeddingmethodscanbedividedintocontinuousembeddingandrandomembeddingmethods.Continuousembeddingmethodreplacestheleastsignificantbitplaneofimage,fromlefttoright,fromtoptobottomorder.Therandomembeddingmethodusesapseudo-randomalgorithmtoselectsomebitsfromtheleastsignificantbitplaneof[50]carrierimagetoreplace.Inthisway,itcanguaranteethesecretinformationrandomlydistributeintheplaneoftheleastsignificantbitofthecarrierimage.Randomembeddinghasgreatersecuritythanthecontinuousembedding.LSBsteganographymethodcanbedividedintotheLSBsubstitutionmethodandLSBmatchingmethod.LSBsubstitutionissubstitutedwiththedesiretoembedsecretinformationoftheleastsignificantbitofthevectorimage.Theupperplaneoftheoriginalimagewiththelowestbit-planeisrepresentativeofthesecret[51]informationtothenewimagecontaininghiddeninformation.Randomembeddingalgorithm,retreatmentofmessagem,(4.8)Replacingleastsignificantbitwiththemessagem′:(4.9)44万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesWhereRep()isfunctionofrandomselectionposition.Randomembeddingalgorithmdistributesthemessagerandomlyintheentirespace,usingthekeytocontroltheembeddingposition.Itssecurityisbetterthanthecontinuousembeddingalgorithm.However,therobustnessispoor.AndmostoftheotheralgorithmsusedtocomparetheLSBalgorithmbiggestadvantagesare:1)Havingalargevolumeofinformation.Accordingtothedifferentrequirementsofthewatermarkimagefidelity,theamountofhiddeninformationbasedonthereplacementof[52]LSBhidingmethodcanreach1-3bits/pixel.Thisinformationhidingcapacityisdifficulttocomparewithotherhiddenalgorithm.2)Thecomplexityofthealgorithm,thecalculationissimple.Accordingtotheknowledgeofthedigitalimage,replacethecarrierpixelsintheairspaceoftheLSB[53]calculationisverysimple.Evenwithrandomintervalstoimproveconfidentiality,theamountofcalculationthanthetransformdomaininformationhidingmethodsignificantlysmaller.ThisisasignificantadvantagetoreplacetheLSBmethod.3)Thewatermarkimagedistortionissmall,thecarrierandtheoriginalcarrierembedsecretinformationbythenakedeyecannotdistinguish.Hide1bitperpixel,thewatermark[54]imagecanguaranteeahighvisualquality.Evenreach2bits/pixel,3bits/pixelhiddenwatermarkimagecanstillensurehighfidelity.4)Thesecretinformationishiddenamongthecontentsofthevectordatainsteadofthefirstclassfileatwhichtopreventfrombeingdestroyedduetoformatconversion.However,theLSBalgorithmpresenceofthemajorlimitationsisthehiddendatapoorrobustness,whenthevectorfileisdamagedorwhenthebycertainproceduresnoiseinterference,inwhichthehiddendataisextractedwillbedifficulttocomplete.Althoughthereareshortcomings,butbecausemostoftheimageinformationhidingtosecretcommunications,theperformanceofthecommunicatingpartiesmostconcernedaboutinthisapplicationare:security,imperceptibility,theamountofinformation,theLSBalgorithmiswellpositionedtomeetthiskindsofapplications,so,asalargeamountofdata,informationhiding,LSBalgorithminahiddencommunicationstilloccupyanimportantpositionto45万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesbecomethemainstreammethodofaninformationhiding.Itisbasedonthealgorithmintheidealenvironment,sothealgorithmishighlydesirable.4.3.3ThemagicmatrixDigitalimagehasthefeaturesofvivid,informativeandeasytotransmissionovertheInternet.Ithasbecomethemainformofthenetworktotransmitinformation.Inordertoprotectthelegitimaterightsoftheownerandpreventinformationleakage,transmissionneedsafe,reliableandlowtimecomplexityimageencryptionalgorithm.Scramblingmethodisanidealmethodtohidethetrueinformationofthedigitalimage.Anditispretreatmentmethodtoimplementotherimageinformationsecurityalgorithms.Itsprincipleisthatchangeeachpixelorlocationinthedigitalimagetogenerateameaninglesssignalandprotecttherealinformationdigitalofimage.Scramblinghastwomethods:periodicandaperiodic.Theformerincludesarrangementofmagicsquare,ArnoldtransformandFASScurves.Thelatterincludeschaostransform,Fouriertransformandconvolutiontransform.Butduetothelimitofthealgorithmofthemagictheretocreated,acertaincorrelationexistsbetweenthetransformimagesofthemagicsquareadjacentpixelgroups,sothatthetransformationoftheimageforsomeiterationweakenedsecurity.Therefore,theauthorproposesthevirtualhologrammagictransformationtohelpenhancesecurityforimageencryption.2Thedefinitionofthemagicsquare:1,2,...,nnaturalelementsofordernmatrix:[](4.10)IftheMelementsatisfies∑∑∑(4.11)ThenyousaidthemagicsquarematrixMisanaturalnumber,referredtoasthemagicsquare.Whereiscalledthemagicsummation.Magicsquareencrypteddigitalimagesolution:amatrixPandthemagicsquarematrixMisselectedaccordingtotheranksofone-to-onecorrespondenceintheimagematrix.46万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesWherein,thesizeofPandMisequal.Melementsai,jandPelementsbi,jconductoperations,obtainedtransformationmatrixM1,theelementsoftheimagematrixPcorrespondingmobilelocationM1.Ltimestransform,inmagicsquare,thearbitraryelementai,jmovetothepositionoftheelementTL(ai,j).()()(){(4.12)()When,()()(4.13)2Thusanarbitraryelementai,jbacktoitsoriginalposition,i.e.niterationstransformation,eachpixelisthenreturnedtothepositionbeforeconversion,sothemagic2transformationcycleoftheimageT=n.4.4EvaluationConcealment,soundnessandsecurityofadigitalwatermarkingalgorithm,alloftheseissuesneedtopracticalandobjectiveevaluationtools.Thereisatradeoffbetweenperceptionandrobustnessofthewatermark.Ifwewanttoassessperceptionofwatermark,wecanmeasurethembyasubjectivetestorqualitymetrics.VarianceMSE(MeanSquareError)usedinthestatisticalprocessisaveryusefulstatisticalcharacteristicsindex.Themeansquareerrorcanbeadirectreflectionofthechangeinassessmentofobjects.Withthemeansquareerror,wecaninsightintoavarietyofbehavioralcharacteristicsoftheevaluationobject.Themeansquareerrorcanbeusedtoestimatechangesofimagequality,formeasurechangesinthequalityoftheoriginalimagewithaddingawatermark.Heregivestheobjectiveindicatorsofthequalityofanimagechange.MeansquareerrorformulaisasshowninEquation:∑∑(4.14)Inaspecificpracticalapplication,itisusedtocalculatethepeaksignal-to-noiseratioatlast.SNRasameasureofthescaleofthereconstructedimagequality,butthecalculationisslightlymoredifficult.ItisgenerallyusingPSNRasameasurefortheordinaryimageformat,47万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesthegraylevelsfrom0to255,with0representingblack,255white,thereforePSNRformulaisasshownfollow:(4.15)Thefourdocumentimagesarelistedbelow;theyareshowingtheeffectofdifferentmethodsofwatermarkinthetestingprocess.Wherein,thefirstpictureistheoriginalimage.Theremainingthreewatermarkimageusingdifferentmethods.Whereintheembeddingstrengthis0.1,theembeddeddatalengthis3000bits,thesameinformationdata.Figure4.4TestoriginalimageFollowingamapLSB,DCT,DWTmethodswatermarkeddocumentimage.Ascanbeseenfromtheimage4.5,TheleftoneisusingLSBmethod,onlythenakedeyeobservation,itisdifficulttoseethedifferencewiththeoriginalimage.ThemiddleoneisobtainedusingtheDCTmethodwatermarkeddocumentimage.Canseethattheupperpartofthetextdoesnotchangemuch,butthelowerpartofthedocumentimagetoproducealotofnoisechanges.TherightoneiswatermarkeddocumentimageobtainedusingtheDWTmethod.Ascanbeseen,inthemiddleofthedocumentNextarealotofdistortionandnoise.48万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesFigure4.5LSB,DCT,DWTwatermarkedimageFromtheabovefigure4.5,wecanseethatusingtheDCTandDWT,thepictureshaveseriousdefacedhappens,graphicwatermarkDCTthanDWTwatermarkgraphictosomeimprovement,butstilltherearesomerecognizable.Inthemethod,LSB,afterembeddingawatermarkandtheresultingimpactisverysmall,therefore,oursolutionismoresuitablefortheuseoftheLSBmethod.Ofcourse,therearesomeclassicwatermarkingmethodswehavenottested.LSBmethodhastheadvantageofalargeramountofinformationhiding,buttheuseofthismethodtoachievethedigitalwatermarkisveryfragileandcannotwithstandthelosslessandlossinformationprocessing,andifknowexactlywatermarkhiddeninseveralLSBThedigitalwatermarkcaneasilybeerasedorbypassed.Butourapproachismodified,joinedhandwritingcharacteristics,soyoucanbetterfixthisproblem.Therefore,wehavechosentousetheLSBdocumentimagewatermarkembedding.Embeddeddataandimagesarelarger;itseffectisquitegood.Thefollowingtable4.1isthepremiseofthesamedataandtheoriginalimage,usingthetypeofmethod,andcomparesthePSNRvaluesobtainedinthedifferentembeddingstrength.ItcanbeseenclearlyfromTable4.1,theLSBmethodisgoodinthisapplication.49万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesTable4.1ComparisonofdifferentmethodsPSNRvalueEmbeddingstrengthLSBDCTDWT0.1PSNR67.442716.23897812.34060.5PSNR67.439436.29930212.39670.9PSNR67.441076.3378412.42461Table4.2belowusingtheLSBmethodPSNRvaluesobtainedbytheembeddedwatermarkcomparison,inthecaseofadifferentamountofdata,thesameoriginalimage.Fromthetableitcanbeseenthatthefeed-offintheprocessofdataincreases,PSNRvaluesdecreased,buttherateofdeclineintheacceptablerange.ThePSNRvalueremainswithinarelativelyhighrange;itispossibletoplayagoodeffect.Table4.2PSNRcomparisonofdifferentdatadatasetsPSNR185.02048281.99313380.21928478.95256577.96588677.18803ThefollowingTable4.3showstheuseofdifferentsizesoftheoriginalimage,thewatermarkcanaccommodatetheamountofdatatest,andgivesthePSNRvalueinthecorrespondingcase.Inthiscase,theembeddingstrengthis0.1.Table4.3ThePSNRvalueunderthecapacitytestSRCimgdataPSNR192kb48kb51.170893.8M220Kb48.97424.8MB3.88MB48.41899InTable4.4test,weselectedthecaseofthetwoextremesofthedocumentimage.Acompletelyblankdocumentimage,theotheroneisfullofallthecharactersinthepositionofdocumentimages,asshowninfigure.Embeddingstrengthselectedhereis0.1,theembeddeddatafor3000-bitinformationdata.50万方数据
Chapter4Multiplayer-documentsWatermarkSignatureWithHandwritingFeaturesTable4.4ExtremedocumentimagetestPSNRSRCblackSRCblankPSNR61.3405161.3393Figure4.6Testfullgridgraph4.5SummaryThisresearchanalyzestheconceptandcharacteristicsofdigitalwatermarking.Thenweresearchtheimagedatastructure.Wehaveproposedadigitalwatermarkingschemebasedonthehandwritingcharacteristics.Wehopethatinpaperless,digitalimageaswatermarkembeddingthecarrier,combinedwithhandwritingfeaturestoensuretheauthenticationofthesignature.Intheprocessofwatermarkembedding,weanalyzethehumanvisualsystemandmakefulluseofavarietyofmaskingcharacteristicsofthehumanvisualsystem.Wedesignthefeatureregionintheimagesuitableforembeddingthewatermarkembeddingalgorithm.Itmakesthewatermarkembeddingisnoteasytohumanvisualperception.Wemayreducetheattackoccurred.Onthebasisofthewatermarkimperceptible,weimprovedthewatermarkrobustness.Wedesignedatypeofwatermarkembeddingstrengthindifferentregions,differentchannels.Inaddition,wealsousedthepretreatmentmethodtoimprovewatermarkrobustnessofthewatermarkimage.Follow-upstudies,wewillcontinuetofindtheproblemandcorrectthetitle.Thesametime,wewillstudythistopiccanreallybeappliedtoreallifetomultiplewatermarks,andmanypeoplesignedcertification.51万方数据
Chapter5TwoStagesDocumentWatermarkingWithFrequencyBasedAlgorithmCHAPTER5TWOSTAGESDOCUMENTWATERMARKINGWITHFREQUENCYBASEDALGORITHM5.1IntroductionandapplicationThemainpurposeofthischapterisproposingatwostagesdocumentwatermarkingmodel.Thesetwostageswatermarksprovideamoresecureenhancementforthedocumentswatermarking.Inthefirststage,thefeatureistheusingofwaveletmethodstoaddanidentificationandpasswordwatermarktobiologicalsignals,suchastheECGandhandwritingsignature.Aftercompletingtheprocess,themodelcanextracttheECGandhandwritingdatatouseridentification.Inoperationofthebeginning,ECGinformationisconvertedintoabinaryform.Intherecognitionphase,wewillsegmentandstretchECGinformationwithoutdestroyingtheimportantinformationinthepremiseofaffectingthejudgment.Inthesecurityresearch,throughbiometrictoprotectpersonalprivacyismightbecomeatrendforthedocumentwatermarking.Forexample,inahospital,patients'medicalrecordsarestoredinadatabasesystem.Butrarelyhospitalstakeasecureapproachtoprotectthepatients’information.Infact,thisisunfairfortheprivacyofpatients.Formanyhospitals,thepatientinformationmighthaveagreatrisk.Ifthesedataarestoredintheirsystemwithaweakprotection,malicioususersandhackerscaneasilyusethemforotherpurposes.Inaddition,ifthedataismodified,thereisnotomanyapproachtoprotectthem.Thenthepatient'srightmightbeaffected.Thescenarioexampleisshowninfigure5.1;firstlyacertificationserialnumberisembeddedintheECG.Thenthedoctorssignonmedicaldocumentsandsavethesignatureimagefile.Afterthat,there-addingwatermarkECGfragmentasawatermarkisaddedtothedocumentimage.Bysuchmeans,itcanplayaprotectiverole.Afterthedocumentimagetransmissionorotheruse,wecanbringoutECGandhandwritingdatawiththepreviouscomparison,ifthesimilarityishighenough,thenwecanconsiderthatthedocumentisnot52万方数据
Chapter5TwoStagesDocumentWatermarkingWithFrequencyBasedAlgorithmdamagedandstolen.Then,wecanextractthewatermarkstoreserialnumberbycomparingtheresultsachieveddoublecertification.Figure5.1Hospitalapplicationdiagram5.2Thenewarchitecture5.2.1BackgroundWiththepresentsituation,therearenotdirectlyrelatedcontentstotheresearchthatwehavedone.UsingECGwatermarkandhandwritingdatatoenhancedocumentsecurityisanewsurvey.However,thereisalreadysomeinformationaboutwatermarks,whichallowsustoreference.Currently,thewatermarkinformationprotectionresearchonECGisstillinitsinfancy,therearesomerelatedstudiesinfigure5.2.ECGsignalwatermark,mainlyusingwavelettransformdigitalwatermarkingencryptiontechnology.Therefore,inthisfieldofresearchisverypromising.53万方数据
Chapter5TwoStagesDocumentWatermarkingWithFrequencyBasedAlgorithmFigure5.2RelatedworksaboutthemodelAccordingtothediagramabovewecansee,biometricwatermarkverificationmainlyincludestwoaspectstechnologyofthewatermarkandverification.Intheaspectofdigitalwatermark,theycanbedividedintotransformdomain,fractal,spatialdomain,cepstrumdomain,spreadspectrumandsoon.Amongthem,wemainlyusedtransformdomainmethodssuchasDCT,FFTandDWTinpreviousresearch.5.2.2ThemodeloftwostageswatermarkingThroughtheaboveexperimentalresults,thisresearchproposedanewarchitecturemodel.Itsdiagramisshowninfigure5.3below.First,themodelcollectstheuser'sECGsignal,andembedstheuserIDnumberasawatermarkintotheECGsignal.Thenthemodelcollectstheuser'shandwrittendata,includingthepressure,speedandothercharacteristicvalue,etc.Afterextractionalgorithmabstracthandwrittendataofcharacteristicvalue,thenfuseECGdatawiththehandwrittendata.Themodelexecuteddatatransformationencryption54万方数据
Chapter5TwoStagesDocumentWatermarkingWithFrequencyBasedAlgorithmoncombineddata,andthenembeddedtheminthedocumentimage.Afterthesesteps,itwillbeuploadedtothereceiveronthenetwork.Ifthedataisn’tdamaged,thereceiverwillextractfuseddatafromtheimage.Thentheycanfurtherextractandanalysisitscontents.Throughthisinformation,theycomparedthemwiththeoriginalinformation,soastodeterminethereliabilityofthedata.Figure5.3ThenewmodeloftwostageswatermarkingThroughtheabovedescription,themodelcontainstwolayersofwatermarking,andbeabletosupportthemulti-playerssignaturefeatures.Sometechnologythismodelusedisbasedonthepreviousexperiments.Firstofall,ECGiscutintofragments.Ineachperiod,weusewaveletdecomposition.Itisbreakdownintosevenlayers.IDnumberorpasswordisembeddedinthecoefficients’bottom.Intablets,handwritteninformationcanbeextracted.HandwritteninformationandECGdataaremixedtogether.Weusethesegment-insertdatafusionmethod.Then,themodelusesmagicsquaretransformationtoencryptdata.WehavetestedthemethodsofDWT,DCT,FFTandLSBinthepreviousexperiments.Finally,themodelusestheKLTmethodtoembedthedataintheimage.Inthisway,theexperimentwouldfindabestmethodtoapplytothismodel.55万方数据
Chapter5TwoStagesDocumentWatermarkingWithFrequencyBasedAlgorithm5.3SummaryInthischapter,wemainlyputforwardanewmodel.Thismodelisbasedonthesignaturefeatureoftwostagesdocumentwatermarkingarchitecture.Thetechnologythisarchitectureuseshasbeendescribedinthepreviouschapter.Butwewanttoinvestigatemoreoptimizationalgorithm.Therefore,inadditiontothepreviouslychosenalgorithm,weselectanewalgorithmforexperiments.Bydifferentexperiments,wehopetofindabetterwayforthemodel.Theexperimentisinprogress.Togetthefinalresults,weneedmoreexperiments.56万方数据
Chapter6ConclusionCHAPTER6CONCLUSIONThisarticlemainlymadeasurveyinthefollowingthreeaspects.Firstofallistheelectrocardiogram(ECG)digitalwatermarking.ThispartmainlyadoptsthemethodofwavelettransformbyembeddingidentityIDtoprotectinformation.Thesecondpartmainlyusesthewavelettransformmethodfordatacompressionandrecognition.ThethirdpartmainlyemploysthemethodofLSBandmagicsquarefordocumentimagewatermarking.Attheendofthearticle,anewarchitecturemodelwasputforward.Themodelisusedtoobtainhumanbiologicalcharacteristicsandprotectinformation.Themaincontributionsofthisarticleareasfollows:(1)Inthefirstpart,accordingtothecharacteristicsofECG,thewatermarkembeddingmethodwasimprovedthroughtheresearch.Watermarkwasembeddedinthesmoothareasothatthesubsequentrecognitionworksbetter.(2)Inthesecondpartofthecompression,accordingtothecharacteristicsofthewaveletcompression,thestudygaveupthehighfrequencycomponenttoreducetheamountofinformationandimprovethecompressionefficiency.(3)Intermsofrecognition,thispapermainlyputforwardacuttingandstretchmethodaccordingtothefeaturesofECG.Afterthat,itextractedcharacterdomainofECG,thusimprovedtheefficiencyofrecognition.(4)Thehandwritingaspectmainlyabstractedthecharacteristicsofhandwrittendata,combiningwiththemagicsquarereplacementandLSBwatermarktostrengthensafetyperformance.(5)Attheendofthethesis,thispaperputsforwardanewtwostagesbasedwatermarkingarchitecture.Thisarchitectureusesseveralmethodswhichhavebeendescribedearlier.Thelistaboveisthebasicworkinthisthesis.Accordingtopreviousresearch,manyresearchescanbestudiedonbiologicalcharacteristicswatermarking.Accordingtodo57万方数据
Chapter6Conclusionresearchesandexperiment,thehandwrittenfeatureswatermarkingalsohasalotoftechnology.Inthefutureresearch,wewillalsocontinuetoexploreandresearch,andtrytoimprovetheframework.58万方数据
ReferencesREFERENCES1.M.S.Nambakhsh,A.Ahmadian,M.Ghavami,R.S.Dilmaghani,andS.Karimi-Fard.AnovelblindwatermarkingofECGsignalsonmedicalimagesusingEZWalgorithm.EngineeringinMedicineandBiologySociety,3274-3277,2006.2.A.Ibaida,I.Khalil,R.V.Schyndel.ALowComplexityHighCapacityECGSignalWatermarkforWearableSensor-netHealthMonitoringSystem.ComputinginCardiology,393-396,2011.3.Y.Zigel.AnalysisbysynthesisECGsignalcompression.ComputersinCardiology,279-282,1997.4.S.Kaur,O.Farooq,R.Singhal,andB.S.Ahuja.DigitalwatermarkingofECGdataforsecurewirelesscommunication.RecentTrendsinInformation,TelecommunicationandComputing,140-144,2010.5.K.M.Zheng,X.Qian.ReversibleDataHidingforElectrocardiogramSignalBasedonWaveletTransforms.ComputationalIntelligenceandSecurity,1:295-299,2008.6.D.Anand,U.C.Niranjan.Watermarkingmedicalimageswithpatientinformation.EngineeringinMedicineandBiologySociety,2:703-706,1998.7.S.M.S.Jalaleddine.ECGdatacompressiontechniques-aunifiedapproach.BiomedicalEngineering,37(4):329–343,1990.8.S.Q.Wu,J.W.Huang,D.R.Huang,Y.Q.Shi.EfficientlySelf-SynchronizedAudioWatermarkingforAssuredAudioDataTransmission.TransactionsonBroadcasting,51(1):3078-3092,2005.9.Y.S.Kim,O.H.Kwon,R.H.Park.Waveletbasedwatermarkingmethodfordigitalimagesusingthehumanvisualsystem.ElectronicsLetters,35(6):466-468,1999.10.X.Kong,S.MemberandR.Feng.WatermarkingMedicalSignalsforTelemedicine.InformationTechnologyinBiomedicine,5(3):195-201,2001.11.M.Engin,O.ÇıdamandE.Z.Engin.Wavelettransformationbasedwatermarkingtechniqueforhumanelectrocardiogram(ECG).JournalofMedicalSystems,29(6):589-594,2005.12.J.W.Huang,Y.Wang,andY.Q.Shi.Ablindaudiowatermarkingalgorithmwithself-synchronization.CircuitsandSystems,3:627–630,2002.13.S.T.Chen,H.N.Huang,J.S.Pan,K.K.Tseng,andS.Y.Tu.Audiowatermarkingquantizationbasedonminimum-lengthscalingscheme.CryptographyandSecurity,1:30,2011.14.Y.L.Kan.DigitalAudioWatermarkTechnologyBasedonFourierTransform.JournalofBeijingBroadcastingInstitute,12(1):62-67,2005.15.R.I.Mohammad.ImplementingofSPIHTandsub-bandenergycompression(SEC)methodontwo-dimensionalECGcompression:anovelapproach.EngineeringinMedicineandBiologySociety,3763-3766,2005.16.X.Y.Wang,H.Y.Yang,H.Zhao.Digitalaudiowatermarkingalgorithmbasedonadaptivequantization.TechnicalAcoustics,23(2):117-120,2004.59万方数据
References17.S.G.Miaou.Wavelet-basedlossy-to-losslessECGcompressioninaunifiedvectorquantizationframework.BiomedicalEngineering,52(3):539-543,2005.18.K.M.Zheng,L.W.Tang,X.Qian.Watermarkingtechnologyforelectrocardiogramsignalcertification.ComputerEngineeringandApplications,45(20):231-233,2009.19.Zigel,Y.ECGsignalcompressionusinganalysisbysynthesiscoding.BiomedicalEngineering,47(10):1308-1316,2000.20.D.Patra,M.K.Das,S.Pradhan.IntegrationofFCM,PCAandneuralnetworksforclassificationofECGarrhythmias.InternationalJournalofComputerScience,36(3):5-9,2010.21.M.Javadi,R.Ebrahimpour,A.Sajedin,S.Faridi,S.Zakernejad.ImprovingECGclassificationaccuracyusinganensembleofneuralnetworkmodules.PLoSONE,6(10),2011.22.S.Parminder,K.M.Preetinder.FastFourierTransformationBasedAudioWatermarkingusingRandomSample.InternationalJournalOfAdvancedEngineeringSciencesAndTechnologies,9:66-68,2011.23.H.H.Chou.AhighperformancecompressionalgorithmforECGwithirregularperiods.BiomedicalCircuitsandSystems,2(4):9-12,2004.24.M.Brito.ApredictiveadaptiveapproachtogenericECGdatacompression.IntelligentSignalProcessing,32-37,2005.25.S.-T.Chen,H.-N.Huang,C.-J.ChenandG.-D.Wu.Energy-proportionbasedschemeforaudiowatermarking.IETProceedingsonSignalProcessing,4(5):576-587,2010.26.AndrewD.Ker.SteganalysisofLSBmatchingingrayscaleimages.SignalProcessingLetters,12(6):441-444,2005.27.DumitrescuS,XiaolinWuandZheWang.DetectionofLSBSteganographyviaSamplePairAnalysis.SignalProcessing,51(7):1995-007,2003.28.RicherP.Steganalysis.Detectinghiddeninformationwithcomputerforensicanalysis.SANS/GIACPracticalAssignmentforGSECCertification,20-26,2003.29.C.C.Chang,Y.H.Chen,Z.H.WangandM.C.Li.Adataembeddingschemebasedonamagicmatrixandwetpapercodes.ComputationalIntelligenceandNaturalComputing,2:303-306,2009.30.C.C.ChangandY.C.Chou.Afragiledigitalimageauthenticationschemeinspiredbywetpapercodes.FundamentaInformaticae,90:17-26,2009.31.J.Mielikainen.LSBmatchingrevisited.SignalProcessingLetters,13(5):285-287,2006.32.H.JiwuandY.Q.Shi.Adaptiveimagewatermarkingschemebasedonvisualmasking.ElectronicsLetters,34(8):748-750,1998.33.H.JiwuandT.Tieniu.AreviewofInvisibleImagewatermarking.ActaAutomaticaSinica,26(5):645-655,2000.34.HyejoungY,HyungwooL,SangjinLandJonginL.Designatedberificationofnoninvertiblewatermark.LectureNotesinComputerScience,2851:338-351,2003.35.S.T.Chen,G.D.Wu,H.N.Huang.Wavelet-domainaudiowatermarkingschemeusingoptimization-basedquantization.IETSignalProcessing,4(6):720-727,2010.60万方数据
References36.B.S.Ko,R.NishimuraandY.Suzuki.Time-spreadechomethodfordigitalaudiowatermarkingusingPNsequence.Acoustics,SpeechandSignalProcessing(ICASSP),2:2001-2004,2002.37.W.Bender,D.GruhlandN.Morimoto.Techniquesfordatahiding.IBMSystemsJournal,35(3/4):313-336,1996.38.M.AcharyyaandM.K.Kundu,Wavelet-basedtexturesegmentationofremotelysensedimages.ImageAnalysisandProcessing,69-74,2001.39.S.BandyopadhyayandS.K.Pal.Pixelclassificationusingvariablestringgeneticalgorithmswithchromosomedifferentiation.GeoscienceandRemoteSensing,39(2):303-308,2001.40.H.Kim.Stochasticmodelbasedaudiowatermarkwhiteningfilterforimprovedetection.Acoustics,SpeechandSignalProcessing(ICASSP),6:1971-1974,2000.41.S.KatzenbeisserandF.A.P.Petitcolas.Informationhidingtechniquesforsteganographyanddigitalwatermarking.Imprimealademande,16-24,2000.42.A.Ysmail.Imagequalitystatisticsandtheiruseinsteganalysisandcompression.TherequirementsforthedegreeofdoctorinBogaziciUniversity,2001.43.D.X.Qi,J.C.Zou,X.Y.Han.Anewclassofscramblingtransformationanditsapplicationintheimageinformationcovering.ScienceinChina,43(3):304-312,2000.44.S.M.MohidulIslam,R.DebnathandS.K.AlamgirHossain.DWTbaseddigitalwatermarkingtechniqueanditsrobustnessonimagerotation,scaling,jpegcompression,croppingandmultiplewatermarking.InformationandCommunicationTechnology,246-249,2007.45.G.Caronni.Assuringownershiprightsfordigitalimages.Proceedingsof‘reliableITsystems’VIS95,251-263,1995.46.A.WestfeldandA.Pfitzmann.Attacksonsteganographicsystems.LecturenotesinComputerScience,1768:61-76,2000.47.S.M.S.Tarzjani,S.Ghaemmaghmi.DetectionofLSBmatchingusinggraylevelrunlengthmatrix.IntelligentInformationHidingandMultimediaSignalProcessing,787-790,2009.48.I.Avcibas,B.SankurandN.Memon.Steganalysisusingimagequalitymetrics.ImageProcessing,12(2):221-229,2003.49.S.H.Liu,H.X.YaoandW.Gao.Steganalysisbasedonwavelettextureanalysisandneuralnetwork.IntelligentControlandAutomation,5:15-19,2004.50.W.N.LieandL.C.Chang.Datahidinginimageswithadaptivenumbersofleastsignificantbitsbasedonthehumanvisualsystem.ImageProcessing,l:286-290,1999.51.N.ProvosandP.Honeyman.Detectingsteganographiccontentontheinternet.NetworkandDistributedSystemSecuritySymposium,1-11,2002.52.P.Neils.Defendingagainststatisticalsteganalysis.USENIXSecuritySymposium,323-335,2001.53.A.SmolaandB.Scholkopf.Atutorialonsupportvectorregression.StatisticsandComputing,14(3):199-222,2004.54.M.P.Queluz,Content-basedintegrityprotectionofdigitalimages.SecurityandWatermarkingofMultimediaContents,3657:85-93,1999.61万方数据
AppendicesAPPENDICESPublishedpapers:1.XialongHe,Kuo-KunTseng,Huang-NanHuang,Shuo-TsungChen,Wavelet-BasedQuantizationWatermarkingforECGSignals,IEEEComputing,Measurement,ControlandSensorNetwork(CMCSN),237-240,2012.7.2.XialongHe,Kuo-KunTseng,ZhaohongChen.《多层次手写特征浮水印融合算法研究》，第十二届离岛资讯技术与应用研讨会论文集，322-326，2013.5。Patents:手写特征与数字文件浮水印融合方法。出版号CN103310403A，申请号CN201310177196，发明者：陈昭宏,曾国坤,张少林,贺夏龙。62万方数据
微信/支付宝扫码支付