基于通道空间注意力机制的视频水印嵌入方法及系统
By adopting a video watermark embedding method based on channel space attention mechanism, the problem of uneven watermark embedding strength caused by the Softmax function is solved, which improves the robustness and readability of the watermark and enhances the model's resistance to attacks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)
- Filing Date
- 2023-07-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning-based video watermarking technologies suffer from watermark embedding strength due to the interdependence of channel probability values caused by the Softmax function, which leads to the loss of some watermark information and reduces the robustness of the watermark.
A video watermark embedding method based on channel spatial attention mechanism is adopted. By extracting spatial and channel feature information of video sequences, a weight transformation module is used to generate a weight matrix, eliminating the influence of the Softmax function on the independence of feature information, and optimizing the watermark embedding position and strength.
It improves the performance of watermark embedding, enhances the robustness and readability of watermarks, improves the training and optimization performance of models, and enhances the ability to resist attacks.
Smart Images
Figure CN116781926B_ABST