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

2026-03-01

AI Summary

Current active watermarking defenses for machine-generated images suffer from insufficient robustness and are vulnerable to low-cost removal attacks. This work proposes Hide&Seek, a general and efficient framework for watermark removal that leverages pixel-level reconstruction combined with lightweight image optimization techniques. Hide&Seek significantly reduces the computational cost of attacks while preserving high visual fidelity. The method effectively removes watermarks across a range of state-of-the-art schemes, systematically exposing the fragility of existing defensive mechanisms and underscoring the urgent need for more robust watermarking approaches.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Algorithmic accountability and transparency on the web

Abstract

Watermarking has emerged as a key defense against the misuse of machine-generated images (MGIs). Yet the robustness of these protections remains underexplored. To reveal the limits of SOTA proactive image watermarking defenses, we propose HIDE&SEEK (HS), a suite of versatile and cost-effective attacks that reliably remove embedded watermarks while preserving high visual fidelity.

Problem

• image watermarking
• machine-generated images
• watermark removal
• robustness
• proactive defense

Innovation

• watermark removal
• pixel-wise reconstruction
• low-cost attack
• image watermarking
• visual fidelity