%0 Journal Article
%T 一种基于自然图像统计模型的图像合成检测方法
%A 孙韶杰
%A 吴琼
%A 李国辉
%J 自动化学报
%P 1564-1567
%D 2009
%R 10.3724/SP.J.1004.2009.01564
%X ？Nowadays,digitalimagescanbeeasilytamperedduetotheavailabilityofpowerfulimageprocessingsoftware.Asdigitalcamerascontinuetoreplacetheiranalogcounterparts,theimportanceofauthenticatingdigitalimages,identifyingtheirsources,anddetectingforgeriesisincreasing.Blindimageforensicsisusedtoanalyzeanimageinthecompleteabsenceofanydigitalwatermarkorsignature.Imagecompositingisthemostcommonformofdigitaltampering.Assumingthatimagecompositingoperationsaffecttheinherentstatisticsoftheimage,weproposeanimagecompositingdetectionmethodonbasedonastatisticalmodelfornaturalimageinthewavelettransformdomain.ThegeneralizedGaussianmodel(GGD)isemployedtodescribethemarginaldistributionofwaveletcoefficientsofimages,andtheparametersofGGDareobtainedusingmaximum-likelihoodestimator.ThestatisticalfeaturesincludeGGDparameters,predictionerror,mean,variance,skewness,andkurtosisateachwaveletdetailsubband.Then,thesefeaturevectorsareusedtodiscriminatebetweennaturalimagesandcompositeimagesusingsupportvectormachine(SVM).Toevaluatetheperformanceofourproposedmethod,wecarriedouttestsontheColumbiaUncompressedImageSplicingDetectionDatasetandanotheradvanceddataset,andachievedadetectionaccuracyof92%and79%,respectively.Thedetectionperformanceofourmethodisbetterthanthatofthemethodusingcameraresponsefunctiononthesamedataset.
%K Imagecompositing
%K generalizedGaussianmodel(GGD)
%K maximum-likelihood(ML)
%K supportvectormachine(SVM)
%K imageforensics
%U http://www.aas.net.cn/CN/abstract/abstract13615.shtml