A video watermarking algorithm based on time factor matrix

Recent studies of digital video watermarking have demonstrated the superiority of the core tensor as a robust feature against various attacks. However, embedding watermarks into the core tensor often introduces perceptible distortions to the host videos. In order to improve the imperceptibility while preserving the robustness, we propose a novel video watermarking algorithm based on a time factor matrix. By applying the Tucker decomposition, we transform the host video into the core tensor and three factor matrices representing the row, column and time axis directions of the video frames, respectively. We found that the first column of the time factor matrix contains the time correlation among frames of the video tensor, which is stable and does not change drastically after geometric deformation or cropping. Hence, our algorithm embeds the watermark by modifying the first column of the time factor matrix. The experimental results show that our algorithm is robust not only to attacks such as cropping, scaling, and rotation, but also to video-specific attacks such as frame deletion and video compression, which outperforms other tensor-based video watermarking algorithms.