A Local Perturbation Generation Method for GAN-Generated Face Anti-Forensics

Although the current generative adversarial networks (GAN)-generated face forensic detectors based on deep neural (DNNs) have achieved considerable performance, they are vulnerable to attacks. In this paper, an effective local perturbation generation method is proposed expose vulnerability of state-of-the-art detectors. The main idea mine fake faces’ areas common concern in multiple-detectors’ decision-making, then generate anti-forensic perturbations by GANs these enhance visual quality and transferability faces. Meanwhile, order improve effect, a double- mask (soft hard mask) strategy three-part loss (the GAN training loss, consisting ensemble classification feature regularization loss) designed for generator. Experiments conducted faces generated StyleGAN demonstrate method’s advantage over methods terms success rate, imperceptibility, transferability. source code available at https://github.com/imagecbj/A-Local-Perturbation-Generation-Method-for-GAN-generated-Face-Anti-forensics .