一种基于深度学习的双JPEG图像压缩检测算法  

A Double Compression Detection Algorithm for JPEG Image Based on Deep Learning

作　　者：楚雪玲 魏为民 华秀茹 李思纤 栗风永 CHU Xueling;WEI Weimin;HUA Xiuru;LI Siqian;Li Fengyong(School of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200090

出　　处：《上海电力大学学报》2020年第5期505-510,共6页Journal of Shanghai University of Electric Power

基　　金：国家自然科学基金(61602295);上海市自然科学基金(16ZR1413100)。

摘　　要：双JPEG图像压缩检测是图像盲取证中的研究热点之一。针对JPEG图像压缩检测盲取证问题,提出了一种在双JPEG格式下基于卷积神经网络(CNN)的压缩检测算法。将图像数据集中的样本以不同的质量因子进行单JPEG压缩和双JPEG压缩,把检测图像的DCT系数直方图作为CNN网络的输入进行特征提取,输出层是样本类别的概率分类。实验结果表明,样本尺寸越大,篡改后的质量因子越大,分类器检测正确率越高;与现有算法相比,提出的算法检测正确率最高提高了1.3%,证明具有良好的双JPEG图像压缩性能检测能力。Double JPEG compression detection is one of the research hotspots in blind forensics.To solve the problem of blind forensics a compression detection algorithm based on convolutional neural network(CNN)in double JPEG format is proposed.The samples in the image data set are compressed by single JPEG and double JPEG with different quality factors,and the DCT coefficient histogram of the detected image is taken as the input of CNN network for feature extraction.The output layer is the probability classification of the sample category.The experimental results show that the larger the sample size is,the larger the quality factor is,and the higher the detection accuracy is.Compared with present algorithms,this algorithm is of the highest accuracy,a rise of1.3%,and has good detection ability of double JPEG compression performance.

关 键 词：双JPEG压缩 图像盲取证 深度学习 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]