双重JPEG压缩图像篡改区域检测与定位

对JPEG(joint photographic experts group)图像实施篡改往往会产生双重JPEG(double JPEG,DJPE) 压缩痕迹,分析该痕迹有助于揭示图像压缩历史并实现篡改区域定位。现有算法在图像尺寸较小和质量因子(quality factor,QF) 较低的时候性能不佳,对两个QF的组合情况存在限制。本文提出了一种端到端的混合QF双重JPEG压缩图像取证网络,命名为DJPEGNet。首先,使用预处理层从图像头文件中提取表征压缩历史信息的量化表 (quantization table,Qtable) 特征,将图像从空域转换至DCT(discrete cosine transform)域构造统计直方图特征。然后,将两个特征输入到由深度可分离卷积和残差结构堆叠而成的主体结构,输出二分类结果。最后,使用滑动窗口算法自动定位篡改区域并绘制概率分布图。实验结果表明,在使用不同Qtable集生成的小尺寸数据集上,DJPEGNet所有指标均优于现有最先进的算法,其中ACC提高了1.78%,TPR提升了2.00%,TNR提升了1.60%。

Tampering region detection and localization of double JPEG compressed images

Tampering with joint photographic experts group (JPEG) images often produces double JPEG (DJPEG) compression traces,and analyzing the traces can help reveal the image compression history and enable tampering region localization.Existing algorithms perform poorly when the image size is small and the quality factor (QF) is low, and there are restrictions on the combination of the two QFs.In this paper,an end-to-end mixed QF DJPEG compressed image forensics network named DJPEGNet is proposed.First,the preprocessing layer is used to extract the quantization table (Qtable) features representing the compression history information from the image header file,and the image is converted from the spatial domain to the discrete cosine transform (DCT) domain to construct statistical histogram features.Then,the two features are input into the main structure formed by stacking the depthwise separable convolution and residual structure,and the binary classification result is output.Finally,a sliding window algorithm is used to automatically locate the tampered region and draw a probability distribution map.The experimental results show that, on small-size datasets generated by different Qtable sets,DJPEGNet outperforms the existing state-of-the-art algorithms in all indicators,with ACC increased by 1.78%,TPR increased by 2.00%,TNR increased by 1.60%.