Skip to content
arXiv· arXiv:2603.01067· Huajie Chen, Tianqing Zhu, Hailin Yang, Yuchen Zhong, Yang Zhang, Hui Sun, Heng Xu, Zuobin Ying, Lihua Yin, Wanlei Zhou·Indexed by this site · Originally published

Hide&Seek: Removing image watermarks at near-zero cost via per-pixel reconstruction

Hide&Seek: Remove Image Watermarks with Negligible Cost via Pixel-wise Reconstruction

AI summary

The researchers propose an image watermark removal attack named HIDE&SEEK (HS), claiming that it can reliably remove embedded watermarks while maintaining relatively high visual fidelity, at very low cost. This work targets current advanced proactive image watermarking protections, aiming to reveal the robustness limits of such protections. The authors describe HS as a set of general and low-cost attack methods for testing the practical effectiveness boundary of machine-generated image watermarking protections.

Read originalarxiv.org