Digital video watermarking algorithm integrating 3D convolution and DT-CWT

With the widespread application of digital video in copyright protection and multimedia security, embedding robust and imperceptible watermarks while preserving video quality remains a critical challenge in digital video applications. Therefore, this study proposes a digital video watermarking algorithm that integrates three-dimensional convolutions with the dual-tree complex wavelet transform (DT-CWT). Unlike existing convolutional neural network DT-CWT watermarking methods, which mainly use DT-CWT as a static frequency domain preprocessing method, this method models DT-CWT sub-band features across frames and spacetime through a three-dimensional convolutional network. Based on the synergistic effects of dense feature fusion and an attention guidance mechanism, it achieves adaptive watermark embedding in the frequency domain and temporal consistency dimensions. A differentiable compression simulation is used for H.264 encoding during training to perceive the actual encoding distortion distribution. This improves the stability of the watermark in real compression environments. The experimental results on the publicly available UVG dataset showed that the proposed algorithm achieved significant performance improvements in multiple indicators. In ablation experiments, the complete model achieved a peak signal-to-noise ratio of 41.8 dB and a structural similarity index of 0.985, validating the importance of key modules. Comparative experiments revealed that after 100 training epochs, the proposed algorithm achieved a bit error rate of 0.22 on the test set while maintaining a recognition rate above 96% under various typical attacks. This method was more resistant to attacks such as scaling, cropping, rotation, and Gaussian noise. The peak signal-to-noise ratio of this method was improved by an average of about 2–4 dB compared to the comparison method. The structural similarity index was improved by about 0.03–0.06. Moreover, the video multi method evaluation fusion index was improved by about 6–12 points. The proposed algorithm significantly enhances the robustness and stability of watermarks while maintaining video quality, providing a high-performance practical solution.