基于张量奇异值分解的鲁棒视频水印算法

针对视频的版权保护, 提出了一种基于张量奇异值分解(T-SVD)的鲁棒视频水印算法. 不同于图像, 视频具有多维度的特点, 能够被看作一个高阶张量, 并通过张量奇异值分解以保留帧内和时域的相关性, 从而提高水印鲁棒性. 首先, 对视频帧的亮度分量进行离散小波变换(DWT), 计算得到低频子带(LL), 并将每3帧的LL作为一个三阶张量; 其次, 将三阶张量分成不重叠块, 并进行张量奇异值分解, 计算得到对角张量; 最后, 利用包含强相关性的对角线张量的奇异值嵌入水印以增强鲁棒性. 此外, 还采用密钥随机提取视频帧, 以加强算法的安全性. 实验结果表明, 提出的算法能够抵抗多种视频攻击, 具有较强的鲁棒性.

Robust video watermarking method based on T-SVD

With a view to better achieve the copyright protection of the video contents, a robust video watermarking method based on Tensor-Singular Value Decomposition (T-SVD) is proposed. In order to preserve strong correlations of the spatiotemporal domain of the video, the video is considered as the high-order tensor, which is transformed using T-SVD to preserve the correlation between the video frames. Firstly, the relationship between the number of video frames, the size of the block, and the embedding capacity is used to determine the number of selected video frames, and to extract the selected video frames. Discrete wavelet transform (DWT) is applied to the luminance components of the selected video frames, and LL sub-bands are extracted. Secondly, LL sub-bands of every three frames are used to construct a third-order tensor. Next, the third-order tensor is divided into non-overlapping blocks, and T-SVD is applied in each block to obtain two orthogonal tensors and one diagonal tensor. Finally, since the singular value of the diagonal tensor includes strong correlations, it can be exploited to embed watermark for robustness. Experiment results shows that the proposed method can resist a variety of image and video attacks.