Research Of Robust Video Watermarking Algorithms And Related Techniques

In recent years, digital watermark is regarded as one of key techniques for some applications, such as copyright protection, content authentication, copy protection, fingerprinting and broadcast monitoring, and this makes digital watermark receive a plenty of attention from a number of researchers. Among the existing watermarking literature, image watermarking algorithms are far more than video's. However, video watermarking algorithms are more useful than image's, and should be received more attentions. Although the existing image watermarking concept can be extended to video watermarking, the video watermarking can not be considered as the simple extension of image watermarking. Video watermarking should be met with some special requirements, such as real-time, random detection and blind extraction. Usually, a perfect video scheme should be at least satisfied with real-time and robustness property simultaneously.In the existing video watermarking literatures, the robustness of video watermarking algorithms usually be neglected in order to meet with real-time property. So, a video watermarking algorithm (so-called mean value reference watermarking, MRW) with real-time and robustness property is proposed. To satisfy the robust requirement, the MRW algorithm adopts three strategies: watermark is embedded into the perceptual significant component, the Watson model is used to control the watermarking strength to maximum and the extended m-sequences is selected as watermark pattern to generate stable mean reference point. In addition, the MRW algorithm is directly performed in the VLC domain and can avoid some complex computations, such as inverse DCT, DCT and motion compensating. Then the MRW algorithm's complexity is very low. However, there still exist some sub-region's energy is far away from the mean value, it will lead to visual quality distortion when the watermark embedding function mandatorily modulates those sub-region's energy to a predefined value. To solve this problem, we propose an optimized mean value watermarking algorithm (OMRW). The OMRW algorithm presents two methods, which are shuffle and decrease detection response value, for this problem specially. Compared with the MRW algorithm, the OMRW algorithm has a better visual quality.However, the MRW and OMRW algorithms'watermark pattern is restricted as the extended m-sequences, which is only identified the user and dose not represent any detailed meaning. Some applications of the video watermarking require watermark pattern to represent some meaningful information. So we propose a differential energy watermarking in the low frequency domain (L-DEW), whose watermarking pattern has not any restrictions. The L-DEW algorithm borrows the energy difference concept from the DEW algorithm and embeds watermark in the low frequency coefficients. Due to enforce an energy difference in the low frequency domain, it must confront with two problems: one is whether it can construct a set of sub-regions with approximate energy; another is how to construct a set of sub-regions with approximate energy. To answer the first problem, we perform a study of statistical model of quantized DCT coefficients. The study results show that there exist a great deal of fragments (i.e. small value coefficient) in the low frequency domain, and this represents it can construct approximate energy sub-regions with low frequency coefficients. To answer the second problem, we perform a research of constructing multi-dimension approximate energy sub-regions. In this research, we present genetic algorithm and improved LPT algorithm. The experiments show that the improved LPT algorithm more suitable for video watermarking. On basis of resolution to the above-mentioned two problems, the L-DEW algorithm first constructs a set of approximate energy sub-regions using the improved LPT algorithm, and then enforces an energy difference between every two sub-regions to embed watermarking bits under the control of the Watson visual model. From the theory analysis and experiments, it is obviously that the L-DEW is indeed better than DEW at the aspect of visual quality, data payload, robustness, and time cost of watermarking embedding.At last, this thesis first models the channel of L-DEW as binary symmetrical channel and then estimates the capacity toward this channel. Then, the RS code and Turbo code, which are suitable for high signal-noise ratio and low signal-noise ratio cases separately, are introduced into the watermarking system for improving the reliability. In the case of L-DEW scheme, Turbo code is better.