Image Resampling Detection Based on Convolutional Neural Network

When an image is under tamper, resampling is one of the most common way to cover the tampering artifacts. With the development of tamper tools, it is difficult to detect the trace of resampling through artificial features to verify the integrity of image. Recently, with the great breakthrough of Deep Learning in computer vision, it is necessary to apply it to the field of digital forensic like resampling detection. As is well known that there is a strongly relationship between the pixels and its surroundings in the resampled image, and the Convolutional Neural Network (CNN) is good at learning the underlying relationship. Low-dimensional feature can hardly find out the trace introduced by resampling while high-dimensional feature is capable of doing this. The CNN has excellent feature extraction ability of distinguishing different feature patterns easily in the high-dimensional space. In this paper, we propose a novel resampling detection supervised CNN that can automatically learn the resampling pattern on the basis of residual mapping relationship. Experimental results show that the proposed method has an excellent performance on different resampling factor detection. Moreover, experiments demonstrate the robustness against the noise of the proposed method. After noise is introduced into the resampled images, our method still learn image resampling pattern and effectively distinguish images with different resampling factors. By detecting the resampled image generated by bilinear interpolation, it is shown that our method is aimed at the resampling pattern rather than the feature pattern of cubic interpolation. Finally, the resampling blind detection experiment show that the proposed CNN can indeed detect the resampling feature pattern.