Robust Image Watermarking Scheme under Halftone Distortion with Surrogate Model
Changsheng Chen, Xijin Li · 2024
Recently, document images have been widely used in various online applications. Digital watermarking is an important forensic technique to verify the authenticity of a document image. However, the recapturing operation leads to a significant risk in document images since the extraction accuracy of recaptured digital watermarking drops significantly. In this work, we propose robust watermark against halftone distortion by utilizing surrogate models and end-to-end watermarking frameworks. Firstly, we employ differentiable surrogate models to generate the halftone distortion. Then, surrogate models are incorporated into end-to-end watermarking frameworks to enhance the robustness of the watermark in the print-camera scenario. To evaluate the robustness of the proposed method, we conduct a series of experiments in real-world scenarios. The experimental results confirm that our method can improve the robustness of the watermark across different devices, datasets, printing sizes, and watermark capacities.