GRRE: Leveraging G-Channel Removed Reconstruction Error for Robust Detection of AI-Generated Images

TL;DR

GRRE addresses the challenge of robust AI-generated image detection under unseen generators by leveraging a G-channel removal strategy. The method removes the green channel to form I_{-G}, reconstructs it with a diffusion-based denoiser to yield \hat{I}_{-G}, and computes GRRE(x0) = \lvert I(x0) - \hat{I}_{-G}(x0) \rvert as the discriminative feature. A ResNet-50 classifier trained on GRRE maps with BCEWithLogitsLoss achieves strong cross-domain generalization and robustness to perturbations. Extensive experiments on the DiffusionForensics DIRE benchmark show superior accuracy and robustness across multiple diffusion and GAN generators, including unseen ones. These results demonstrate the potential of channel-removal-based reconstruction as a practical forensic tool against rapid AI-generated image synthesis.

Abstract

The rapid progress of generative models, particularly diffusion models and GANs, has greatly increased the difficulty of distinguishing synthetic images from real ones. Although numerous detection methods have been proposed, their accuracy often degrades when applied to images generated by novel or unseen generative models, highlighting the challenge of achieving strong generalization. To address this challenge, we introduce a novel detection paradigm based on channel removal reconstruction. Specifically, we observe that when the green (G) channel is removed from real images and reconstructed, the resulting reconstruction errors differ significantly from those of AI-generated images. Building upon this insight, we propose G-channel Removed Reconstruction Error (GRRE), a simple yet effective method that exploits this discrepancy for robust AI-generated image detection. Extensive experiments demonstrate that GRRE consistently achieves high detection accuracy across multiple generative models, including those unseen during training. Compared with existing approaches, GRRE not only maintains strong robustness against various perturbations and post-processing operations but also exhibits superior cross-model generalization. These results highlight the potential of channel-removal-based reconstruction as a powerful forensic tool for safeguarding image authenticity in the era of generative AI.

Figures (3)

• Figure 1: Overview of the proposed G-channel Removed Reconstruction Error (GRRE) framework. Given an input image $x$, the Green-channel Removal Model (GRM) removes the green channel to obtain $x_{-g}$. The diffusion-based reconstruction model generates a restored image $x_{-g}'$, and their difference forms the $x_{-g\_rec}$ map. This map captures channel-sensitive reconstruction discrepancies, which are subsequently leveraged for distinguishing real and AI-generated images.
• Figure 2: Detection robustness of GRRE and baselines under unseen perturbations, measured by AUC and AP across JPEG compression, Gaussian blur, and Gaussian noise.
• Figure 3: Visual comparison among DIRE, GRRE, RRRE, and BRRE on real and generated samples.