面向开放场景的自适应阈值源相机识别算法

打开文本图片集

关键词：图像取证；源相机识别；光响应非均质性；峰值相关能量；孪生网络中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-032-3778-07doi：10. 19734/j. issn. 1001-3695.2025.03.0098

Adaptive threshold-based source camera identification algorithm in open-set scenario

Fan Xingzhena，Liu Yufeiʰ，Tian Huaweia，Xiao Yanhuibt （a.Scloffdkitoftltple'sbutUsit China）

Abstract：PRNU noise servesasadistinctive stochasticfingerprintofimage sensors forsourcecamera identification.Inopenset scenario，theexisting identification methods employafixed threshold tocomparethe magnitudeof PCE，which are incapableofadaptingtofluctuatingchangesofPCEcausedbythecomplexityofimagecontent，image size，andthenumberof images utilized forPRNUfingerprint extraction，resulting inmisjudgmentormisseddetection.Toaddressthisissue，thepaperpro posedanadaptivethreshold-basedsourcecameraidentificationalgorithminopen-setscenario.Specificaly，itdesigned the training modelofadaptivethresholdidentification module，whichcouldtakethePRNUfingerprint，queryimageandthecorresponding noiseresidual，PCEvalueasthesupervisedlearningdata.Propertraining yielded theweightparametersofthemodule，andthen itcouldsetthresholdsadaptively toacommodatePCEfluctuationsandachievemoreaccurateidentification. Multiple sets of comparison experiments with four source camera identification methods on Daxing smart phone dataset and Dresdencameradatasetfullyprovetheefectivenessanduniversalityof theproposedalgorithm，which ismoresuitable for source camera identification tasks with different levels of complexity.

Keywords：image forensics；sourcecameraidentification;photo-response non-uniformity（PRNU）；peak tocorrelationenergy（PCE）；siamese network

0 引言

互联网技术的不断发展与相机、智能手机等成像设备的广泛应用，很大程度上提高了图像获取与传输的便利性。（剩余18326字）