Research On The Key Technologies For Wyner-Ziv Video Coding

The conventional video coding standards, such as MPEG-x or the H.26xrecommendations, mainly depend on the encoder to exploit the correlation statistics of thesource signal to achieve compression. The encoder is typically5to10times more complexthan the decoder due to heavy computations in motion estimation and compensation. Thisasymmetry in complexity suits well application scenarios where a video signal needs to becompressed once but decoded many times, e.g. broadcast or streaming video on demand. Inrecent years, the emerging applications, such as wireless sensor network and wireless videosurveillance systems, bread new challenges to the coding structure characterized by asignificantly lower complex encoder and a higher complex video decoder.Wyner-Ziv (WZ) video coding is a particular case of distributed video coding, a novelvideo coding paradigm based on the Slepian-Wolf and Wyner-Ziv theorems which mainlyexploits the source correlation at the decoder and not only at the encoder as in traditionalpredictive video coding. Therefore, this new coding paradigm may provide a flexibleallocation of complexity between the encoder and the decoder and in-built channel errorrobustness. These benefits make the WZ video codec (WZVC) be used to implement various‘uplink’ video applications in resource-constrained mobile devices, therefore, research on WZvideo coding has great theoretical significance and practical value.Although some progresses have been made in the last few years, the rate-distortionperformance of WZ video coding is still far from the maximum performance attained withpredictive video coding. The WZ video coding compression efficiency depends critically onthe quality of the side information created at the decoder and the capability to model thecorrelation noise between the original information and its corresponding side information.The development of advanced side information creation algorithms and realistic and powerfulcorrelation noise modeling techniques is, therefore, crucial to reach practical and efficient WZvideo coding solutions. In addition, to also address application scenarios where a feedbackchannel is not available, it is necessary to develop encoder driven rate control strategies.In this context, this dissertation focuses on researching and analyzing several keytechnologies of WZ video coding, and the main contributions and innovation points of thethesis are listed as follows:1) A novel framework is proposed to generate side information at the block level in two modes. At first, low compexity coding mode decision at the encoder is performed to classifythe blocks in the WZ frame to Skip mode, Intra mode and WZ mode. For the Skip mode,substitutes the co-located block in the reference frames for side information. While the WZmode corresponds to motion compensated temporal interpolation (MCTI) technique, the Intramode corresponds to a motion compensated quality enhancement technique where a lowquality Intra block sent by the encoder is used to generate the side information by performingmotion estimation with the help of the reference frames. Finaly, the proposed algorithm isintegrated into the proposed WZ video coding framework. Experimental results demonstratethat the proposed schemes can achieve up to2.1dB improvement in rate distortion (RD)performance when compared to state-of-the-art distributed video coding, especially for highmotion video sequences and long GOP sizes.2) A novel pixel-domain correlation noise modeling for WZ video coding is proposed. Inorder to adapt to the video content and coding parameters, we firstly distinguish blocks intotwo cases: high quality side information creation and low quality side information creation,according to temporal and spatial correlation measures. For the high quality side informationcreation case, we model the correlation noise as Laplacian distribution while modeling theother case as Cauchy distribution. Moreover, the location and scale parameter estimation forCauchy distribution is based on quantile estimators. Experimental results show that significantRD improvements of the proposed novel correlation noise modeling are achieved especiallyfor high motion video sequences and large GOP sizes.3) An efficient encoder rate control strategy for a transform domain WZ video codingarchitecture is proposed. The proposed algorithm first obtains an estimate of the sideinformation at the encoder through low complexity side information creation technique, thenchooses the best coding mode in terms of the video content, and then establishes an accuraterate-distortion function model which based on Cauchy distribution to estimate the optimalWZ rate. Finally, the proposed algorithm is integrated into the WZ video coding architecture.Simulations indicate that the proposed Cauchy-based rate allocation algorithm is much moreaccurate than the Laplacian-based method. Moreover, the proposed encoder rate controlsolution achieves a RD performance similar to the decoder rate control solution.