Contrast Adjustment Forensics Based on Second-Order Statistical and Deep Learning Features

Abstract

Contrast adjustment (CA) is one of the most common digital image retouching methods. However, it is often used to hide the traces of tampering, and thus has become an indirect evidence of image forgery. Therefore, CA blind detection has attracted widespread attention in the field of image forensics in recent years. Considering forensic methods based on first-order statistical features are vulnerable to encountering anti-forensic and other operation attacks, an second-order statistical-based CA forensics method using improved Gray-Level Co-occurrence Matrix (GLCM) network is proposed. Different from conventional CNN, which usually takes the image as input, this method can convert input images of different resolution into a uniform size GLCM matrix, in which GLCM adds four more directions to the traditional direction, and then learns the distribution features from GLCM through CNN layers and classifies them. Through active learning the hierarchical feature representation and optimizing the classification results, the proposed network is more suitable for detecting contrast adjustment tampering. Experimental results show that the proposed method can not only detect traditional CA, but also detect the CA image of anti-forensic attacks, and its performance is better than conventional forensic methods.