Research On Video Watermarking Technique Based On H.264/SVC

Digital Multimedia becomes more and more insecure because it is easy to be modified. Traditional encryption technology can no longer meet the needs. Watermark technology provides an effective solution. With the extensive use of video service, the copyright protection of video has become increasingly prominent. Based on the summary of the existing digital video watermark algorithms, this thesis focuses on video watermark techniques for the latest video coding H.264/SVC. The main work is summarized as follows.First, a digital watermark algorithm based on DFT and log-polar mapping for H.264/SVC is proposed. At first, algorithm does 2-d DFT on every picture, then resampling the Fourier magnitudes into log-polar coordinates, and summing a function of thess magnitudes in log-polar coordinates along the log-radius axis, and then forms a one-dimensional signal vector. The watermark is embed into the one-dimensional vector, and apply the inverse log-polar mapping followed by the inverse Fourier transform, then we can get the watermarked video image. In log-polar coordinates, Scaling of the image results in amplification of the one-dimensional signal. This can be eliminated by a correlation detection technology. So the algorithm is robust to scaling attack. Experimental result demonstrate that the proposed approach is a suitable watermark algorithm for H.264/SVC.Second, combined with H.264/SVC codec, a novel watermark algorithm based on the sign of DCT coefficients for H.264/SVC is proposed. At first, choose the 4x4 DCT blocks for embedding watermark bits, then one bit watermark information is embedded by modify the sign of DCT coefficients, which makes it not depend on the level of them. When embed the watermark, a human vision model for 4x4 DCT blocks in H.264 compression is proposed, the watermark bits would be embedded into the most appropriate position in 4x4 DCT block. And when adjust the AC coefficients, we consider not only the residual DCT coefficients, but also the predicted signal DCT coefficients, this will yield better robustness against recoding attacks. Experiments result demonstrate that this algorithm takes care of robustness and transparency requirements. This algorithm can be used for copyright protection.