Research On Key Technologies Of Distributed Video Coding

With the fast development of multimedia technologies, new multimedia application scenarios are continuously emerging. Like video sensor network, wireless video surveillance, smart phone video recording, multi-view and3D video capturing, these new video application scenarios have posed new challenges on video compression and transmission techniques. The typical feature of these new applications is that there are a large number of video capturing devices, whereas their computing ability, memory spaces, and energy resources are strictly limited. The state-of-the-art video coding standards, such as H.26x and MPEG series, remove the data redundancy through complex calculations at the encoder side. These schemes require high computing ability encoders. Furthermore, in multi-view video coding scenario, in order to achieve ideal compression efficiency by removing inter-view redundancy, tremendous energy and bandwidth have to be spent to transmit the original video data of all views to a central encoder to perform joint encoding. Therefore, traditional video coding schemes are not well suited for these new applications. Whereas distributed coding theories point out that for correlated sources, separate encoding and joint decoding scheme can achieve the same or similar compression efficiency as traditional joint encoding and joint decoding scheme. This feature can well satisfy the low encoding complexity requirement of the new applications. Thus distributed video coding (DVC) is attracting widespread interest and becoming a hot research topic in recent years.Oriented to new video applications represented by video sensor networks, this dissertation studies low encoding complexity distributed video coding techniques for both single-view and multi-view videos. The main contents and novelties of the dissertation are as follows:1. This dissertation proposes a low complexity distributed video coding scheme using trellis coded quantization (TCQ).Based on the distributed video coding scheme with small block Slepian-Wolf coding, this work studies how to bring in high efficiency trellis coded quantization, and combined it with Slepian-Wolf coding and entropy coding. Based on the features of this distributed video coding scheme, an appropriate type of TCQ is chosen and modified accordingly. And the coding efficiency of the distributed video coding scheme is increased with very small amount increase of encoding complexity.2. This dissertation proposes a low complexity distributed multi-view video coding scheme based on the fusion of multiple side information.This work studies the side information generation method in distributed multi-view video coding scheme. Based on the analysis of the special, temporal, and inter-view dependency of multi-view videos, a novel side information generation method is proposed. Inter-view motion prediction is performed based on the space structure correlation, and is fused with the side information based on intra-view spatial and temporal correlation. The quality of side information is improved, and thus the coding performance of the distributed multi-view video coding scheme is improved.3. This dissertation quantitatively analyzes the causes of the large performance gap between distributed video coding and traditional video coding.Currently there is still a large gap of compression performance between distributed video coding schemes and traditional video coding schemes. This dissertation studies the causes of this performance from multiple perspectives, including distributed coding theory, the implementation of distributed source coding, and the implementation of distributed video coding. And through distributed source coding experiments, and exchange the two core modules, side information generation and Wyner-Ziv source coding, in distributed video coding scheme with the corresponding modules, motion estimation and entropy coding, in traditional video coding scheme, quantitative analysis result is given. Finally both theoretical analysis and experiments shows, the large performance gap between current distributed video coding and traditional video coding is caused by multiple causes, in which every cause takes a considerable percentage of the performance impact, instead of simply caused by one dominative key reason.In conclusion, oriented to new video capturing applications presented by video sensor network, with limited encoding ability and resources, this dissertation investigates various source correlation characteristics video signal, and proposes several coding methods to improve the distributed video coding efficiency with low encoding complexity, and quantitatively analyzed the causes of the performance gap between distributed video coding and traditional video coding. The results obtained are of both theoretical significance and practical value.