Residual Transferability in Neural Image Watermarking
Neural watermark residuals can be transferred to forge marks on unrelated images; CoverLock reduces that risk.
Neural image watermarks can be forged by extracting watermark-bearing residuals from released images and transferring them onto unrelated content. The authors define residual transferability and find that architectural design, rather than common training-side changes, accounts for large differences across systems. Two mechanisms that bind watermark evidence to the cover image reduce that transfer. For systems that cannot be redesigned, their plug-and-play CoverLock method improves the security-robustness trade-off versus handcrafted and learned classifier defenses.
- Residual transferability measures watermark evidence that survives transfer onto unrelated images.
- Architecture, not common training variations, explains large differences across watermarkers.
- Two cover-dependence mechanisms suppress transferable watermark residuals.
- CoverLock improves the security-robustness trade-off without redesigning existing systems.
From the original article
1 min read at huggingface.co
Neural image watermarks can be forged by extracting watermark-bearing residuals from released images and transferring them to unrelated content. While prior work has demonstrated this vulnerability, what makes these residuals transferable remains poorly understood.
By Ziping Dong, Qi Li, Xinchao Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.