Multiple-Operation Image Anti-Forensics with WGAN-GP Framework

Jianyuan Wu, Zheng Wang, Hui Zeng, Xiangui Kang

Abstract: A challenging task in the field of multimedia security involves concealing or eliminating the traces left by a chain of multiple manipulating operations, i.e., multiple-operation anti-forensics in short. However, the existing anti-forensic works concentrate on one specific manipulation, referred as single-operation anti-forensics. In this work, we propose using the improved Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) to model image anti-forensics as an image-to-image translation problem and obtain the optimized anti-forensic models of multiple-operation. The experimental results demonstrate that our multiple-operation anti-forensic scheme successfully deceives the state-of-the-art forensic algorithms without significantly degrading the quality of the image, and even enhancing quality in most cases. To our best knowledge, this is the first attempt to explore the problem of multiple-operation anti-forensics.