A Deep Learning Approach to Detection of Warping Forgery in Images.

In recent years, image forensics has received full attention from researchers. A large number of algorithms for image smoothing, JPEG compression, copy-move, and shear tampering were published. However, there are still many image tampering algorithms that are not involved. In this paper, we publish a dataset of image warping, which contains more than 10000 images, and propose a novel convolutional neural network called DWF-CNN to identify warped images. In experiments, we compared the performance with 4 alternative networks. The proposed network with the preprocessing layer of the SRM layer and Bayar convolutional layer got the best result, which reached to the accuracy of 99.36%. The experiments also showed that the network with the regular convolutional layer performed even worse than a random guess. It illustrates the importance of the well-designed preprocessing layer in this research area again.