Prediction mode based H.265/HEVC video watermarking resisting re-compression attack

A Compressed-Domain Robust Video Watermarking Against Recompression Attack

“…Considerable achievements have been made in many works [15], [18], [19], [22], [23], [28], [29], [32], [35], [36] on this problem. Mansouri et al [15] used number of non-zero quantization coefficients (NNZ) to determine the embedding area of the watermark, and adjusts the effect of the watermark on video quality through a priority matrix.…”

“…Yang et al [22] also embedded the watermark by modifying the intra prediction mode. However, these algorithms [19], [22] cannot maintain robustness when QP changes greatly since the intra prediction mode is mostly altered to next prediction mode which may alter after recompression attack [23].…”

“…Gaj et al [36] embedded the watermark in the block with less motion information in the I frame, and modify the 4 × 4 discrete cosine transform (DCT) coefficient to change the NNZ difference in the consecutive intra predicted frames to embed watermark. Gaj et al [23] embedded the watermark by changing the intra prediction mode of the video, by grouping of the intra prediction modes such a way that the mode change due to recompression can be closed within a group, to resist recompression attack.…”

“…During the process of recompression, the change of QP will change compressed domain parameters, therefore, it is difficult to find the locations where the watermark embeds, in further to extract the watermark correctly. In [22], [23], [28], [29], [32], [35], [36], in order to locate these watermark in the compressed video watermarking schemes, they embedded the watermark in the specified locations. In these framework, since the detector or the decoder may not find the same locations due to the changes after recompression or imposed attacks, they used a location map or an embedding position template by private key describing embedding location which is used to find the specified locations where the watermark embeds.…”

“…In practical applications, the video watermarking scheme can resist recompression attack when QP increases greatly have more urgent needs and practical significance. The existing compressed domain watermarking schemes are that they are fragile against recompression attack when QP increases substantially [15], [18], [19] or they have security issue since they need location map when extract watermark [22], [23], [28], [29], [32], [35], [36]. In this article, we try to to realize blind extraction without location map to improve the security of the video watermark scheme as well as it has good performance on robustness to recompression attack when QP increases greatly.…”