The Research On Side Information Generation For Distributed Multi-view Video Coding

The encoding complexity of conventional video coding technologies is much higher thandecoding complexity, because correlations are generally exploited at the encoder side. WhileDistributed Video Coding (DVC) is a new concept, at the encoder side several sources with thesame distribution are encoded independently but at the decoded side they are decoded jointly. DVCapplies to the occasions which need low power consumption, low encoder complexity and strongerror resilience. Thus, in wireless video terminator such as wireless video monitoring system andvideo sensor network DVC is widely used. Side information plays an important role in DVC.Distributed Multi view Video Coding (DMVC) can use not only temporal side information, but alsointer view side information. Therefore, the paper studies the generation method of side informationin DMVC and proposes improved methods. Specific work includes the following aspects:The paper first does the research about temporal side information of DVC. By comparing withseveral traditional temporal side information prediction methods, A new temporal side informationalgorithm is proposed based on adaptive weighed and optimized motion estimation interpolationmethod. This method combines adaptive weighted frame interpolation and adaptive search windowalgorithm. It calculates the time correlation between WZ frame and adjacent key frame. Thendifferent weights are used for frame interpolation according to disparate correlation in decoding end.The method can achieve more precise side information by adjusting search range of motionestimation. The proposed method has good performance in both subjective and objective quality.Secondly, the generation method of inter view side information is studied in detail. Bycomparing with several typical inter view side information generation method, the paper researchestrifocal tensor based on simulated annealing algorithm to generate inter view side information. Themethod is proved correct and effective by the experimental results.Finally, in the previous research results foundation, the paper discusses side information fusionmethod based on the motion objects separated from background, including motion object extraction,side information fusion of motion object from both temporal and inter view direction, the fusion ofmotion object and background. The experimental results demonstrate the effectiveness of themethod proposed in the paper.