Research On Reconstruction In Distributed Video Coding

Distributed video coding (DVC) has developed for about10years since2002. Unlikethe traditional video codec with complex encoder and simple decoder, the inter correlation isnot considered, but independent encoding and channel coding techniques are utilized in DVCencoder, While the inter correlation and channel decoding techniques are applied in decoder.Such structure reduces the encoder complexity. Thus, DVC is more suitable for the wirelessvideo applications where computing power, storage capacity and power consumption arelimited.This paper firstly researches on the fundamental theory and framework of DVC. Thenthe paper focuses on reconstruction module and proposes improvement reconstructionmethods in pixel domain DVC and DCT domain DVC respectively. The main work andachievements are as follows:1. Introduce the basic principles and two typical system frameworks of DVC: pixel andDCT domain DVC frameworks. And then compare the DVC compression performance withthe H.264/AVC intra-frame coding. The results illustrate that comparing to H.264Intra-framecoding, DVC has better performance for low-motion-complexity video sequences, while notso well for high-motion-complexity video. Improving the compression performance forhigh-motion-complexity video is now an important research field for DVC.2. According to the characteristics of the reconstruction module in the pixel domaindistributed video coding, Bi-direction Motion Compensation (BMCR) Reconstructionscheme is proposed. In this scheme, the pixel will be reconstructed as its side informationvalue directly if its side information value falls in the correct quantization bin. Otherwise, itsforward and backward compensation values are first calculated by bi-direction motioncompensation. Then the compensation values are corrected and averaged to obtain pixel’sreconstruction value. Experimental results show that compared with the commonly usedMinimum Mean Squared Error (MMSE) reconstruction scheme, the proposed BMCR schemegets an average of about0.5dB in RD performance and improves the visual quality of thedecoded video.3. In order to eliminate the block artifacts in DCT domain DVC, based on the advanced adaptive deblocking filter in H.264/AVC, an improved Boundary Analysis method with MeanAbsolute Differences (MAD) is proposed. This method utilizes the MAD between sideinformation and reconstructed frame as the measure of boundary strength. The thresholds todistinguish different filter mode are obtained by extensive experiments. Experimental resultsillustrate that the algorithm can elevate the performance of DCT domain DVC up to0.7dB,effectively reduce the block artifacts and improve the visual quality of decoded video at lowbit rate.