关键词: 关键词: 视频取证, 水印, 屏摄攻击, 深度学习, 鲁棒性

Abstract: Abstract: Recently, with the advancement of multi-source shooting devices and sensing technologies, screen recording has become a primary channel for video information leakage during dissemination. The screen-capture process causes irreversible physical damage to video watermark information, making it difficult to trace the source of information leaks, which poses a significant challenge in the field of video forensics. Existing anti-screen-capture video watermarking techniques are still in their preliminary stages, primarily relying on handcrafted frequency-domain features, which struggle to effectively resist cross-device screen-capture attacks. To address this issue, this paper proposes an end-to-end anti-screen-capture video watermarking method based on a spatiotemporal feature enhancement network, capable of resisting screen-capture attacks across devices. First, to enrich watermark embedding features, a spatiotemporal feature enhancement network module is designed by integrating an optical flow estimation network and a high-pass filter, optimizing the watermark embedding strategy. Then, by simulating key factors of screen-capture channel noise to replicate screen-capture distortions, an anti-screen-capture attack noise layer is designed to enhance cross-device robustness against such attacks. Finally, validation in real screen-capture attack scenarios demonstrates that the proposed method achieves a decoding accuracy of over 95%, with both robustness and image quality outperforming existing solutions.

Key words: Keywords: video forensics, watermarking, screen-capture attack, deep learning, robustness