Research On Error Resilient Side Information And Correlation Noise Model For Distributed Multiview Video Coding

In recent years, multiview video systems have become more and more popular due to the adoption of the interactive multimedia applications such as 3D television, surveillance and wireless sensor networks. Multiview Video Coding (MVC), as core technology of multiview video system, also has been invested many years. According to the balance of coding complexity, MVC could be divided into two categories:conventional MVC that is used for applications with complex encoder and simple decoder, and Distributed MVC (DMVC) which is suitable for scenario with simple encoder but complex decoder. This paper focuses on DMVC and proposes new algorothms for its error resilience and effective compression.Firstly, the theory of DMVC and Distributed Video Coding (DVC), and basic framework of DVC are reviewed in this paper, for these two codecs share the same theorization and coding structure. Besides, the differences of DMVC and DVC are also introuduced. Based on these differences this paper try to generate robust and effecive DMVC.Secondly, this paper implements high efficiency DMVC framework base on DISCOVER codec and novel algorithms for DMVC such as view side information, side information fusion and correlation noise model. This framework could guatantee that the proposed algorithms compare perfromance with latest methods on the same platform.Thirdly, this paper introduces the error resilience side information for DMVC. Although DMVC has intrinsic error resilient capbality due to channel coding, it could not effectively elimate side information corruption when temporal key frames drop packets. Therefore, this paper designs the robust side information on the basis of multiview motion estimation for temporal key frames corruption scenario. Experiemental results show that the poposed side information achieves much better quality compared with others.Finally, characteristic of correlation noise in DMVC is analized in this paper. According to analysis results, because of view correlaiton, traditional correlation nosie model proposed for DVC could not describe correlation noise in DMVC any more. According to fusion scheme, this paper proposes novel correlation noise models for pixel domain DMVC and DCT domain DMVC respectively. Experimental results prove that the novel algorithms could model correlation noise accurately.