Research On The Corresponding Processing Technology For Multiview Video Coding

With the booming of 3-D application, multiview videos come into the academic sight with its high application value. But it is a huge challenge to deal with the multiview videos because of its large number of views and huge data volume. This paper covers some relevant research of multiview video coding, including random access evaluation of multiview video coding prediction structure, fast algorithms of multiview video coding, compressed sensing-based depth map compression, and color correction of multiview videos.Random access performance is essential for the prediction structure of multiview video coding. To accurately measure the random access performance, a new evaluation method of random access is proposed. This method is based on graph theory, which treats the prediction structure of multiview video coding as a direct acyclic graph, and establishes a mathematical model of the prediction structure. Based on this model, analysis of the dependences of frames in a prediction structure is made, and the concept and calculation method of random access degree is proposed. This method can objectively evaluate the random access performance of the prediction structure of multiview video coding simply and effectively.Multiview video requires a higher speed of coding process. Two fast algorithms for multiview video coding are proposed: Adaptive Search Region Adjustment (ASRA) and Adaptive Selection of Inter-view References (ASIR). ASRA uses the statistical property between the disparity vector and global disparity vector to dynamically adjust the search range, and ASIR uses the prediction information of the lower temporal layer in the Hierarchical B Picture Prediction Structure to reduce the unnecessary number of inter-view references of pictures in the higher temporal layer. These fast algorithms are mainly used in the inter-view prediction of multiview video coding, and can be integrated with other traditional fast algorithms proposed for intra and inter prediction.There exists color discrepancies among different views, which will affect the prediction efficiency of video coding. A de-correlated color correction algorithm for multiview video coding is proposed. It adopts the means of correlation analysis and distance analysis to get the best matching points, and according to the matching point value of tri-stimulus of color, a linear irrelevant color space-based de-correlated correction algorithm is designed. The results show that the algorithm is effective for color correction of multiview videos, and the subjective impressions of the color among different views are identical, high bitrate-distortion gain would be obtained by encoding the corrected videos.It is feasible to combine compressed sensing with image and video coding technology. A new point of view is proposed that quantization is sampling or measurement in compressed sensing. After quantizing the coefficients of 2-D DCT and integer transform, a reconstruction method of image with these coefficients is studied based on compressed sensing, and the encoding and decoding process is designed correspondingly. This method is particularly effective for images with sparse gradient, and the depth maps, which are widely used in 3-D image and video view synthesis just have this kind of feature. Therefore, the compressed sensing is applied to the coding of depth maps, and the method is tested and combined with the practical coding systems as JPEG and H.264/AVC. The results show that compressed sensing can greatly improve the coding efficiency of the depth map.