CycleGAN without checkerboard artifacts for counter-forensics of fake-image detection

TL;DR

This paper introduces a new CycleGAN model that eliminates checkerboard artifacts, enhancing the ability to counter-forensics in fake-image detection by addressing artifacts common in GAN-generated images.

Contribution

It presents the first CycleGAN architecture designed without checkerboard artifacts, improving fake-image detection and counter-forensics.

Findings

CycleGAN without checkerboard artifacts effectively counters forgery detection methods.

The proposed model reduces artifacts that compromise fake-image detection.

Enhanced robustness of fake-image detection against GAN-generated manipulations.

Abstract

In this paper, we propose a novel CycleGAN without checkerboard artifacts for counter-forensics of fake-image detection. Recent rapid advances in image manipulation tools and deep image synthesis techniques, such as Generative Adversarial Networks (GANs) have easily generated fake images, so detecting manipulated images has become an urgent issue. Most state-of-the-art forgery detection methods assume that images include checkerboard artifacts which are generated by using DNNs. Accordingly, we propose a novel CycleGAN without any checkerboard artifacts for counter-forensics of fake-mage detection methods for the first time, as an example of GANs without checkerboard artifacts.