Research on Image Tampering Detection and Localization Based on Iterative GAN

Image tampering detection and localization are important tasks in digital image forensics. Its purpose is to identify and locate the modified or forged regions in the image. With the development of image processing technology, the possibilities of image manipulation are becoming more and more diversified and hidden, which poses challenges for conventional methods based on manual feature extraction. This paper proposes research on image tampering detection and localization based on an Iterative generative adversarial network (Iterative GAN). First, a fake-to-true transform generator GT is designed to generate fake images consistent with the actual image style according to image style transformation. Secondly, a positioning generator GM based on a multi-decoder single-task strategy is proposed to eliminate the artifacts caused by the generator GT, which is used to predict the mask and confidence of tampered regions. Thirdly, according to the characteristics of the iterative generative adversarial network, a loss function that comprehensively considers the feedback of the generator and discriminator is constructed to accurately describe the problem of image tampering detection and localization. Finally, an image tampering detection and localization system based on the iterative generative adversarial network is constructed, and the effectiveness of the proposed method is proved by simulation and experimental verification.