Anti-Forensics of JPEG Compression Using Generative Adversarial Networks

Yingmin Luo, Hanqi Zi, Qiong Zhang, Xiangui Kang

Abstract: JPEG compression is one of the most popular image compression methods. The manipulation history of JPEG compression provides evidence of operational information about the device or software utilized to generate an image. Besides, JPEG compression introduces complex compression artifacts, particularly the blocking artifacts, ringing effects and blurring, which lower the image visual quality. Therefore, erasing the traces left by JPEG compression is of great importance. To solve this problem, we present a JPEG compression anti-forensic method adopting the framework of generative adversarial networks (GANs). This architecture consists of a generator and a discriminator, where the generator can automatically learn how to hide the traces left by JPEG compression during the optimization process against the discriminator. Through extensive experiments, it is demonstrated that the anti-forensically modified images generated by our method can deceive the existing JPEG compression detectors and have very good visual quality.