Universal Image Manipulation Detection using Deep Siamese Convolutional Neural Network

Detection of different types of image editing operations carried out on an image is an important problem in image forensics. It gives the information about the processing history of an image, and also can expose forgeries present in an image. There have been few methods proposed to detect different types of image editing operations in a single framework. However, all the operations have to be known a priori in the training phase. But, in real-forensics scenarios it may not be possible to know about the editing operations carried out on an image. To solve this problem, we propose a novel deep learning-based method which can differentiate between different types of image editing operations. The proposed method classifies image patches in a pair-wise fashion as either similarly or differently processed using a deep siamese neural network. Once the network learns feature that can discriminate between different image editing operations, it can differentiate between different image editing operations not present in the training stage. The experimental results show the efficacy of the proposed method in detecting/discriminating different image editing operations.