文章编号：1002-2082(2024)02-0337-09基于光响应非均匀性的WhatsApp压缩视频来源识别陈懿辉，田妮莉，潘晴，苏开清（广东工业大学信息工程学院，广东广州510006）摘要：光响应非均匀噪声(photoresponsenonuniformity，PRNU)是光学成像传感器成像时引入的一种独特噪声，可有效识别压缩视频的来源。针对现有算法提取压缩视频的PRNU效果并不显著的问题，论文提出了一种改进PRNU提取算法。首先，去除视频编解码的环路滤波器，对视频帧使用双密度双树复小波变换进行分解；然后对高频子带使用基于贝叶斯阈值估计的双变量收缩算法进行估计，再使用自适应加窗维纳滤波进行二次估计，得到噪声残差；最后用基于量化参数值加权的最大似然估计法聚合噪声残差，再与视频帧估计得到PRNU。实验结果表明：该文提出的方法在20s时WhatsApp视频的识别率为75%。关键词：光响应非均匀性；源相机识别；压缩视频；双密度双树复小波变换；双变量收缩中图分类号：TP391文献标志码：ADOI：10.5768/JAO202445.0201009WhatsAppcompressedvideosourcecameraidentificationbasedonphotoresponsenonuniformityCHENYihui，TIANNili，PANQing，SUKaiqing（SchoolofInformationEngineering,GuangdongUniversityofTechnology,Guangzhou510006,China）Abstract：Photoresponsenonuniformity(PRNU)noiseisauniquenoiseintroducedtoopticalimagingsensorsduringimagingandcanbeeffectivelyappliedtothesourcecameraidentificationofcompressedvideo.DuetotheproblemthatexistingalgorithmsdonotproducesignificanteffectonextractingPRNUofcompressedvideo,animprovedalgorithmtoextractPRNUwasproposed.Firstly,theloopfilterofvideocodecwasremoved,andthevideoframewasdecomposedbydoubledensity-dualtree-complexwavelettransform.Then,thehighfrequencysubbandwasestimatedbybivariateshrinkagealgorithmbasedonBayesianthresholdestimation,andtheadaptivewindowWienerfilterwasusedforsecondaryestimation.Finally,afterthenoiseresidualswereobtained,theywereaggregatedbythemaximumlikelihoodestimationmethodbasedonquantizationparameterweighting,andthePRNUwasestimatedwithvideoframes.ExperimentsontheVISIONdatasetshowthattheaccuracyoftheproposedPRNUextractionmethodinWhatsAppcompressedvideorecognitionisimprovedto75%at20s.Keywords：photoresponsenonuniformity...
