Detection of Image Manipulations Using Siamese Convolutional Neural Networks

Abstract

The processing history of an image can reveal the application of different types of image editing/manipulation operations applied to images and also can expose forgeries. This paper proposes a novel deep learning-based manipulation detection method using a siamese neural network. The advantage of the proposed method is that it can even detect manipulations not present in the training stage. The network is first trained to differentiate between different types of image editing operations. Once the network learns feature that can discriminate different image editing operations present in the training stage, the unknown manipulations are detected using the one-shot classification strategy. We show that the network can also check whether an image is downloaded from a social media platform or not. The experimental results validate the efficacy of the proposed method.