Robust GAN-Face Detection Based on Dual-Channel CNN Network
Yong Ming Fu, Tanfeng Sun, Xinghao Jiang, Ke Xu, Peisong He · 2019
Nowadays, the identification of Generative adversarial networks (GAN) generated face images has become an important issue. Unfortunately, existing methods cannot detect these images with post-processing operations efficiently, such as image denoising and image sharpening. In this paper, the images are pre-processed by a gaussian low-pass filter, the combination of pre-processed images and the high-frequent components of original images can mitigate the influence of various image contents and can improve the detection capability against some widely-used image post-processing operations. Therefore, we carefully design a dual-channel structure based on Convolutional Neural Network (CNN) aiming to extract robust representations for detection of GAN generated face images. Extensive experiments are conducted on the public available dataset. Experimental results demonstrate that the proposed approach outperforms the state-of-the-arts with several image post-processing operations.