摘要
光响应非均质性(PRNU)噪声具有唯一性和稳定性,可作为成像设备的指纹用于源相机识别。现有的识别方法在开放场景下采用固定阈值比较峰值相关能量(PCE)大小,无法适应图像内容复杂度、图像尺寸、用于PRNU指纹提取的图像数量等带来的PCE波动变化,导致了误判或漏检。针对此问题,提出了一种面向开放场景的自适应阈值源相机识别算法。具体地,设计了自适应阈值识别模块的训练模型,可将PRNU指纹、查询图像及其噪声残差、PCE值等作为监督学习的数据。通过适当训练得到模块的权重参数,进而自适应设定阈值以适应PCE值的波动、实现更精准的识别。在Daxing智能手机数据集和Dresden相机数据集上与四种源相机识别方法的多组对比实验充分证明了所提算法的有效性和普适性,更能适应不同内容复杂程度的源相机识别任务。
PRNU noise serves as a distinctive stochastic fingerprint of image sensors for source camera identification.In open-set scenario,the existing identification methods employ a fixed threshold to compare the magnitude of PCE,which are incapable of adapting to fluctuating changes of PCE caused by the complexity of image content,image size,and the number of images utilized for PRNU fingerprint extraction,resulting in misjudgment or missed detection.To address this issue,the paper proposed an adaptive threshold-based source camera identification algorithm in open-set scenario.Specifically,it designed the training model of adaptive threshold identification module,which could take the PRNU fingerprint,query image and the corresponding noise residual,PCE value as the supervised learning data.Proper training yielded the weight parameters of the mo-dule,and then it could set thresholds adaptively to accommodate PCE fluctuations and achieve more accurate identification.Multiple sets of comparison experiments with four source camera identification methods on Daxing smart phone dataset and Dresden camera dataset fully prove the effectiveness and universality of the proposed algorithm,which is more suitable for source camera identification tasks with different levels of complexity.
作者
范兴振
刘宇飞
田华伟
肖延辉
Fan Xingzhen;Liu Yufei;Tian Huawei;Xiao Yanhui(School of Information and Network Security,People’s Public Security University of China,Beijing 100038,China;School of National Security,People’s Public Security University of China,Beijing 100038,China)
出处
《计算机应用研究》
北大核心
2025年第12期3778-3784,共7页
Application Research of Computers
基金
公安部科技计划技术研究项目(2022JSYJC22)。
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
image forensics
source camera identification
photo-response non-uniformity(PRNU)
peak to correlation energy(PCE)
siamese network
