SimuFreeMark: A Noise-Simulation-Free Robust Watermarking Against Image Editing

TL;DR

SimuFreeMark introduces a robust image watermarking method that embeds watermarks into low-frequency components using a pre-trained VAE, eliminating the need for noise simulation and enhancing resistance to various attacks.

Contribution

This work presents a noise-simulation-free watermarking framework that leverages low-frequency stability and deep feature embedding, advancing robustness against diverse image editing attacks.

Findings

Outperforms state-of-the-art methods under various attacks

Eliminates the need for noise simulation during training

Maintains high visual quality of watermarked images

Abstract

The advancement of artificial intelligence generated content (AIGC) has created a pressing need for robust image watermarking that can withstand both conventional signal processing and novel semantic editing attacks. Current deep learning-based methods rely on training with hand-crafted noise simulation layers, which inherently limit their generalization to unforeseen distortions. In this work, we propose $\textbf{SimuFreeMark}$, a noise-$\underline{\text{simu}}$lation-$\underline{\text{free}}$ water$\underline{\text{mark}}$ing framework that circumvents this limitation by exploiting the inherent stability of image low-frequency components. We first systematically establish that low-frequency components exhibit significant robustness against a wide range of attacks. Building on this foundation, SimuFreeMark embeds watermarks directly into the deep feature space of the low-frequency components, leveraging a pre-trained variational autoencoder (VAE) to bind the watermark with structurally stable image representations. This design completely eliminates the need for noise simulation during training. Extensive experiments demonstrate that SimuFreeMark outperforms state-of-the-art methods across a wide range of conventional and semantic attacks, while maintaining superior visual quality.