Study On Watermark For MPEG Codec

With the development and popularization of digital TV and Internet multimedia technology, the copyright protection becomes more important. Digital watermarking is one of a candidate for effective protection. The technique of watermark for still images is matured.However, video watermarking still attracts many researchers in recent years. New algorithms of video watermarking based on MPEG-2 and AVS is studied in this thesis. Three novel methods are proposed after deeply studying the H&G arithmetic and the motion estimation and the motion compensation algorithms in MPEG-2.1) A novel method of watermarking is proposed by utilizing the characteristics of the concavity and convexity of function in frequency domain. In this method, the video is decoded partly, when the picture is I frame, then entropy decoded it, and get the quantized DCT coefficients. Three middle frequency coefficients are selected in each block. Watermark is embedded in these coefficients by the characteristics of the concavity and convexity of function. In this method, watermarking is robust to Gauss noise and quantization attack, and also to salt and pepper noise.2) A new method of drift compensation is proposed. This method is less complex compared with the H&G arithmetic. In this method the computation power of entropy decoding, inverse quantization, inverse DCT transformation and motion compensation are dramatically reduced comparing to the H&G arithmetic. All blocks of picture are drift compensated in the H&G arithmetic, in this method only a few blocks which meet certain condition are drift compensated. The quality of the picture of the P frame and B frame is improved after compensation3) An improved algorithm is proposed based on the H&G algorithm. In the proposed algorithm the watermark signal is high pass filtered before embedded into the video signal. Because only the part of high frequency embedded into video signal, the invisibility of the watermark is improved. Besides this, it also makes smaller affection to the bit rate. This method can also be used in the algorithm of the watermarking in the frequency domain, because the watermark can be high pass filtered in both spatial domain and frequency domain.All three algorithms have been evaluated in the video streams encoded by the MPEG-2 and AVS codec. the platform of Visual C++ and Matlab 6.5 were selected for the simulation experiment.