Improving GAN-Generated Image Detection Generalization Using Unsupervised Domain Adaptation
Mingxu Zhang, Hongxia Wang, Peisong He, Asad Malik, Hanqing Liu
Abstract
In recent years, with the significant improvement of Gener-ative Adversarial Networks (GANs), fake images generated by GAN become hardly distinguishable from real ones, thus threatening the authentication of digital images. To resolve this issue, several fake image detectors based on supervised binary classification have been designed. However, current methods remain vulnerable when testing samples are gener-ated by an unknown GAN model. In this work, an unsuper-vised domain adaptation strategy is introduced to improve the performance in the generalization of GAN-generated image detection by using a small number of unlabeled images from the target domain. Self-Attention block and novel loss function have been constructed to optimize the domain adaptation process, thus getting a better generalization. Experimental results demonstrate that the proposed scheme achieves high detection accuracy with few unlabeled images in the target domain, which shows that unsupervised methods can be used for the detection of GAN-generated images.