Reproduce “Practical Deep Dispersed Watermarking with Synchronization and Fusion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hengchang Guo, Qilong Zhang, Junwei Luo, Feng Guo, Wenbin Zhang, Xiaodong Su, Minglei Li
ByteDance researchers developed DWSF, a deep watermarking framework enabling practical, robust embedding and extraction for arbitrary-resolution images. It achieves superior visual quality with 47.38dB PSNR, minimal file size increases, and outperforms state-of-the-art methods by an average of 23-40% in bit accuracy against challenging combined attacks.
Key insights
- Dispersed, block-based embedding with a consistent message allows the watermarking framework to scale to arbitrary image resolutions while maintaining high visual quality and minimal file size overhead.
- Combining a dedicated segmentation model (U2-Net), a Pad&Split strategy, and geometric transformation estimation provides robust synchronization crucial for extracting watermarks from images subjected to geometric attacks.
- Similarity-based message fusion from multiple embedded blocks significantly enhances the reliability of watermark extraction by filtering unreliable messages, making the system highly robust against various distortions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.