关键词: 数字水印, 版权保护, 内容认证, 视频恢复, 数字音视频编解码标准

Abstract:

Firstly, a non-negative matrix factorization with sparseness constraints on the parts (NMFSCP) method is proposed, which is used to extract the video movement component. The AC energy coefficients of DCT transformation are used as the texture feature simultaneously. Then the linear combination of two video features above is used to adjust the location and strength of the watermark adaptively. Secondly, one dynamic robust watermark and two fragile watermarks are generated according to the energy of the I-frame macro-block, which are used to detect and restore the attack position respectively. Finally, the robust watermark and two fragile watermarks are embedded into the medium and high frequency DCT coefficients which are the biggest of the luminance blocks in the I-frame respectively. Simulation results show that the embedded robust watermark is robust not only to general attacks but also to video special attacks, and that the mixture recovery detection mechanism based on two types of fragile watermarking is sensitive to tamper, has good accuracy of tamper localization and can recover the video frame location under the time-resynchronization attacks. The video stream need not be decoded completely for watermark extraction in this paper, so it meets the needs of real-time detection.

Key words: watermarking, copyright protection, content authentication, video recovery, audio video coding standard