A fast motion segmentation based watermarking for MPEG-2 video

To achieve high robustness against collusion and geometric attacks, video watermarking schemes often employ local-content-based methods in which watermarks are embedded in several semantically independent regions that partitioned from video-segments. For these schemes, self-synchronization is also an important issue that needs to be taken into account in the design of watermark detectors. In this paper, a simple and fast object-based watermarking algorithm is proposed. It is based on the techniques of motion segmentation and block matching, which is consistent with MPEG-2 standard. In the proposed algorithm, a spread spectra watermark that similar to a CDMA signal is inserted into selected video blocks so that different video contents can be protected accordingly. Experimental results have demonstrated the advantage of the scheme over other exist local content-based ones in terms of computational complexity, resistance against attacks include cropping, pixel shift and recompression etc.