| With the rapid development of computer, multimedia and network techniques, digital products have been popularized increasingly and can be easily acquired. However, it also results in the problem that digital products can be easily copied, pirated and tampered under no control. The research of digital watermarking is to embed the copyright information into the digital products without visible influences, and thus to protect the digital products.Video watermarking is one of the most difficult and active area in the digital watermarking research. Compared with image watermark, the video watermarking algorithm has to satisfy the blind detection and real-time properties in addition to the basic robustness and imperceptibility requirements, and that makes the existing image watermarking techniques could not be used in the video watermarking system directly. This thesis focuses on the research of the imperceptibility and robustness of video watermark, and the main work and contributions are as follows:Firstly, Human Visual System (HVS) properties are researched, and an adaptive video watermarking algorithm with high imperceptibility is proposed based on Watson Visual model and the characteristics of MPEG video data. In the algorithm, the basic properties of HVS are utilized for watermark placement, and the Watson model is modified to fit the structure of MPEG video stream, besides, multi-frame detection method is used to improve the detection results. Experimental results show that the algorithm is with high imperceptibility and good robustness, and can resist to JPEG compression.Secondly, different kinds of methods against geometric attacks are analyzed, and a geometric invariant algorithm based on DCT feature blocks is proposed which can satisfy the real-time embedding requirement. In the algorithm, certain amount of DCT feature blocks are selected as the embedding points, and the synchronization block is calculated accordingly for frame registration before detection. On the detection side, the feature blocks are first located for the calculation of the synchronization block, and then neighborhood search is implemented on the synchronization block to accurately estimate the geometric distortion. After frame registration, feature blocks are extracted again for the final watermark detection. Experimental results show that the watermarking system can resist to geometric attacks well, with satisfying good invisibility, real-time and security properties.The research in the thesis is valuable both in theoretical and practical aspects. The two proposed video watermarking algorithms satisfy blind detection and real-time properties, thus can be used in the copyright protection of video content. |
