ARIW-Framework: Adaptive Robust Iterative Watermarking Framework

TL;DR

This work tackles copyright protection for images generated by large models by addressing the trade-off between visual quality and robustness in watermarking. It introduces the Adaptive Robust Iterative Watermarking Framework (ARIW-Framework), which learns a robust residual $R$ added to the original image so that $X_1 = X + R$, while embedding strength is guided by image gradients and attack-specific robustness weights. The encoder employs parallel noise simulation to optimize robustness across multiple attacks, the decoder aggregates multi-layer outputs to reliably extract the watermark, and the loss combines image fidelity and watermark accuracy with dynamic weights for each attack. Experiments across diverse datasets show high visual quality (PSNR/SSIM) and robust watermark recovery under various distortions, with strong generalization and supportive ablation findings that validate the architectural choices.

Abstract

With the rapid rise of large models, copyright protection for generated image content has become a critical security challenge. Although deep learning watermarking techniques offer an effective solution for digital image copyright protection, they still face limitations in terms of visual quality, robustness and generalization. To address these issues, this paper proposes an adaptive robust iterative watermarking framework (ARIW-Framework) that achieves high-quality watermarked images while maintaining exceptional robustness and generalization performance. Specifically, we introduce an iterative approach to optimize the encoder for generating robust residuals. The encoder incorporates noise layers and a decoder to compute robustness weights for residuals under various noise attacks. By employing a parallel optimization strategy, the framework enhances robustness against multiple types of noise attacks. Furthermore, we leverage image gradients to determine the embedding strength at each pixel location, significantly improving the visual quality of the watermarked images. Extensive experiments demonstrate that the proposed method achieves superior visual quality while exhibiting remarkable robustness and generalization against noise attacks.

Figures (4)

• Figure 1: (a) Existing deep learning watermarking framework improves the robustness by employing serial noise simulation within the noise layer $N$. (b) Proposed framework in this work enhances the robustness by employing parallel noise simulation within the encoder $E$.
• Figure 2: The pipeline of our framework.
• Figure 3: Watermarked image $X_1$ and residual image $R$ generated under different embedding strengths.
• Figure 4: Convergence of PSNR and SSIM corresponding to different ablation experiments.