Wyner-Ziv Bayer-pattern video coding

This thesis addresses the problem of Bayer-pattern video communications using Wyner-Ziv video coding (WZVC). It is studied how to take advantage of the low encoding complexity and the inherent error resiliency of WZVC for video compression and transmission. Moreover, it is investigated how to optimize the coding process specifically for Bayer-pattern video in order to improve the coding efficiency. To begin with, a state-of-the-art Wyner-Ziv video codec using turbo codes is optimized and its functionality is extended. Firstly, the turbo decoding is optimized by making good use of error-free parity bits, which leads to a considerable improvement of coding efficiency. Secondly, a comparative study of bit-plane based coding using binary turbo codes and symbol based coding using trellis merging in turbo decoding is carried out. Experimental results show that the latter one is considerably more error-resilient than the former one when the transmission channel is quite noisy. Thirdly, motion estimation is distributed between encoder and decoder flexibly. In this way, motion estimation is made scalable at the encoder. Moreover, error-resilient video transmission over an AWGN channel using WZVC is studied. This study involves statistical modeling of the sufficient coding rates, estimation of decoding failure probability, end-to-end distortion estimation and joint source-channel rate control using adaptive quantization. For joint source-channel coding using WZVC, two alternative solutions are compared. The first one is to use WZVC for both source and channel coding, while the second one confines WZVC to source coding and carries out channel coding at a separate stage. As a result, the former one is significantly more error-resilient than the latter one. Therefore, WZVC is a promising candidate solution for joint source-channel coding of video. Furthermore, the color space transform, specifically the chroma subsampling, is optimized specifically for Bayer-pattern video. Two color space transform methods, namely B-4:2:0 and B-4:2:2 are proposed. Chroma sample positions are more reasonably chosen than the nominal sample positions in the H.264/AVC standard. Therefore, the proposed methods lead to a significant improvement of the reconstructed RGB video quality. Particularly, the method B-4:2:2 keeps only half of the luma pixels compared to the method B-4:2:2, and thus reduces