Study On Side Information Generation In Distributed Video Coding

In the conventional video coding architecture, encoder undertakes complex computation for motion estimation and motion compensation, the bulk of computing burden is located in the encoder, and the decoder is uncomplicated comparatively, the architecture is suitable for the situation for which video has to be compressed once but decoded repeatedly, when it comes to the application such as mobile video communication and wireless video surveillance system, the computation power, battery volume and communication bandwidth are limited, low complexity encoder is admirable, Distributed Video Coding system, or DVC for short, is proposed consequently. DVC performs intra-frame encoding and inter-frame decoding and exploits statistic correlation of frames in the decoder, the task of motion estimation and motion compensation is shifted from encoder to decoder smartly, theoretically speaking, the compression efficiency matches that of conventional video coding, furthermore, DVC is laid on the basis of channel encoding, thus inherently capable of error resilience, in conclusion, DVC architecture achieves the goal of compression storage and robust transmission for video data.DVC is a new paradigm in the area of video compression which involves other technology domain, and theoretical research innovation on side information generation and rate control technique plays an critical role for practical application, especially improvement on side information generation algorithm, the more accurate side information is generated, the fewer parity bits are needed in the decoder and the better rate-distortion system performs.In this thesis, we introduce theoretical foundation and related techniques of DVC in the first place, then elaborate the mechanism of an integrated DVC system, subsequently, side information generation algorithms have been studied vastly, and five novel side information generation technique are adopted on the basis of classic algorithms:(1)in the searching process of motion estimation, the distortion of labour blocks is considered to optimize the matching criterion; (2)weighted vector-median filter is utilized to correct inaccurate motion vectors; (3)a motion compensation scheme based on multiple frames is adopted; (4)the information extracted from direction of motion vectors is utilized to proceed motion analysis on moving objects; (5)a variable block based hierarchical motion estimation is utilized to obtain the motion vectors for the boundary of irregular moving objects.Finally, simulation experiment is proceeded in Matlab environment, experiment results indicate that the algorithms and schemes adopted in this paper lead to better subjective and objective performance compared with classic side information generation algorithms.