Reproduce “OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xuanyu Zhang, Zecheng Tang, Zhipei Xu, Runyi Li, Youmin Xu, Bin Chen, Feng Gao, Jian Zhang

Researchers from Peking University developed OmniGuard, a hybrid deep image watermarking framework that combines proactive embedding with passive blind extraction for enhanced digital image forensics. This method achieves F1-scores over 0.95 for tamper localization and maintains robust copyright recovery against AI-generated content (AIGC) edits, while also improving visual fidelity to over 42dB PSNR.

Key insights

  • Decoupling localized watermark reconstruction precision from tamper localization accuracy allows the embedding network to generate higher-fidelity container images.
  • A passive, degradation-aware tamper extractor, utilizing both the tampered image and reconstructed watermark, can robustly distinguish true tampering from degradation artifacts.
  • Lightweight simulation of AIGC global (VAE compression) and local (partial removal) editing during training effectively enhances copyright watermark robustness against advanced AI manipulations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.