基于多阶段渐进式神经网络的图像相机指纹提取算法  

Image Camera Fingerprint Extraction Algorithm Based on MPRNet

作　　者：高昌锋 肖延辉[1] 田华伟[1] GAO Changfeng;XIAO Yanhui;TIAN Huawei(School of National Security,People's Public Security University of China,Beijing 100038,China)

机构地区：[1]中国人民公安大学国家安全学院,北京100038

出　　处：《信息网络安全》2022年第10期15-23,共9页Netinfo Security

基　　金：国家自然科学基金[61972405];中央高校基本科研业务费专项资金[2022JKF02003]

摘　　要：光响应非均质性(Photo-Response Non-Uniformity,PRNU)噪声的唯一性和稳定性决定了其可作为数码相机的指纹并可用于数字图像的溯源与取证。为了提高PRNU指纹质量及图像溯源精确度,文章提出一种基于多阶段渐进式神经网络的图像相机指纹提取算法。该神经网络使用编码器-解码器架构来学习上下文特征,并利用高分辨率分支来挖掘局部特征,同时在每个阶段引入注意力模块来重新加权局部特征。这样既可以融合上下文信息,又能准确获取局部信息,进而可以充分挖掘图像自然噪声中潜在的PRNU指纹。在Daxing智能手机数据集和Dresden相机数据集上与其他算法的对比实验证明了文章提出的图像相机指纹提取算法的优势。The uniqueness and stability of Photo-Response Non-Uniformity(PRNU)noise determine can be used as a digital camera fingerprint and for traceability and forensics of digital images. In order to improve the quality of PRNU and the accuracy of image traceability,this paper proposed a PRNU camera fingerprint extraction algorithm based on multi-stage progressive image restoration network. The network uses an encoder-decoder architecture to learn contextual features and utilizes a high-resolution branch to mine local features, while introducing an attention module at each stage to reweigh local features. In this way, the global information can be fused, and the local information can be accurately obtained, so that the potential PRNU fingerprints in the natural noise of the image can be fully mined. The comparative experiments with other algorithms on the Daxing smartphone dataset and the Dresden camera dataset demonstrate the effectiveness of proposed PRNU algorithm.

关 键 词：图像处理 图像取证 光响应非均质性 源相机识别 卷积神经网络 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]