From Forgery to Authenticity: Image Anti-Forensics via Reconstruction and Artefact Elimination

Jerry Jiang, Dongliang Chang, Shengjie Chen, Lei Chen, Jiwen Lu, Jie Zhou

Keywords: Image anti-forensics, computer vision

TL;DR: The paper proposes a two-phase image anti-forensics approach to eliminate artefacts in manipulated images, effectively evading existing forgery detectors.

Abstract:

In recent years, the development of large-scale vision-language models has resulted in significant advancements in image generation and editing, producing results that can often deceive the naked eye.
However, despite their convincing appearance, these generated images remain susceptible to detection by forgery detectors due to various artefacts.
The goal of image anti-forensics is to eliminate such artefacts, ensuring that manipulated images successfully evade detection and enhance their overall quality. Existing image anti-forensics methods primarily focus on rectifying artefacts at the feature level, often overlooking the authenticity of the manipulated regions.
To address this limitation, we propose a two-phase approach. In the first phase, we introduce GUIded Diffusive rEfinement (GUIDE), a zero-shot learning-based image refinement module aimed at reconstructing details from unaltered regions.
In the second phase, we introduce an artefact removal algorithm to eliminate artefacts from the reconstructed ''forged regions''. We validate the effectiveness of our proposed method across multiple image forgery datasets, and comprehensive ablation studies further affirm the efficacy of each component of our approach. The code will be made available upon acceptance.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4383