Research On Multi - View Video Coding Predictive Structure

Traditional 2D video can’t provide user scene stereo feeling and satisfy the people to the needs of stereo video viewing.With the development of video coding technology,people begin to study the network adapatability,user interaction and reality of video coding technology, such as multiview video coding technology.Multiview video obtained by multiple cameras in a number of different spatial locations to shoot the same scene.It can provide to users to watch video content in different perspective and stereo feeling.But,the multiview video contains a huge amount of data with increasing the number of viewpoint.How to efficiently compress the video data is becoming the hot topic in the research field of multiview video coding.Prediction structure is the important technique for enhancing efficiency of multiview video coding, thus it is important to optimze the prediction structure.At present, the general structure using hierarchical B pictures is defined in JMVC(Joint Multiview Video Coding) software which is released by JVT(Joint Video Team) organization.Prediction structure of JMVC not only eliminates the spatial redundancy and time redundancy, and eliminates the redundant information among viewpoints by using the disparity estimation technique.Although this prediction structure achieves higher coding efficiency, but also has the poor performance of random access and high coding complexity.How to achieve a balance between coding efficiency and random access by improving the prediction structure become the key of further study.This paper has carried on deep research on prediction structure of multiview coding,and focus on improving the selection method of reference frame selection method and point of view between the reference view.It realises a well balance between coding efficiency and random access.Specific include:1.In the MVC standard, the reference predicting structure uses the fixed architecture which can not adapt to the charateristics of video series.To solve this problem,this paper proposed an improved method of selection of prediction structure I viewpoint.Firstly,analysis the SSIM(Structural Similarity) between the view image and other view image, then select viewpoint which has the maximum SSIM with the other viewpoint as a reference point of view.The experimental results show that, the proposed method can improve the multiview video coding efficiency and the random access at the same time.2.The MVC prediction structure using hierarchical B pictures can not to take advantage of the properties of video sequences which have different temporal and spatial correlation.In order to improve random access performance of the prediction structure,proposed a prediction structure using method of selective discarding based on the improved selection method of I viewpoint..Based on research of temporal and inter-view correlation of mulitview video picture,cancel inter-view prediction of some temporal level(TLID=3 and TLID=4).Experimen-tal results show that the proposed method can get better performance in random access while maintaining coding efficiency.In summary,two kinds of improved prediction structure proposed by this paper have better performance on efficiency and random access than the reference prediction structure of JMVC.It has certain reference significance for multiview video coding and other video coding applications.