A method of H.264 video watermarking robust to attack on I and P frames by removal

The authors consider two new attacks on watermarked video specific to HD 60 fps H.264 High and Main profiles: selective I/P frames' removal and zeroing of residuals in B frames. We propose the method of embedding of a readable watermark for H.264 High Profile video robust to such attacks. The video stream is partially decoded removing entropy coding, then a watermark is embedded into low-frequency DCT residuals and motion vectors of B frames, so that it could be later extracted in an uncompressed domain even if almost all I and P frames were removed by the attacker during transcoding. The method is nonblind, robust to new attacks and to such attacks as transcoding, down-scale, and post processing filters due to robustness of low frequency DCT coefficients. The method consists of two stages: computationally intensive preprocessing stage performed to each video and low complexity payload embedding stage performed to each copy. So it is specifically suitable for fingerprinting of Video-on-Demand (VoD) systems, where a unique watermark should be embedded in each video's copy.