Research On High Quality And Efficient 3D-HEVC Coding

Multi-view plus depth video contains depth information and has become an emerging medium that can offer a richer visual experience than traditional video. An arbitrary three-dimensional scene within certain perspective can be easily constructed by multi-view video. Nowadays, Multi-view plus depth video becomes an efficient three-dimensional video format. It can be used in free-view television, augmented reality and natural user interface. But its application highly depends on the video quality optimizing and efficient coding. Nowadays, the JCT-3V group is committed to 3D video coding standardization. And it has relesed the test platform HTM. Although the 3D-HEVC has high compression performance, its coding efficiency and quality need to be improved. This thesis depends on the HTM to research the key technologies of 3D-HEVC, redesign and improve the most wildly used neighboring block disparity vector derivation method and depth map intra coding, to fulfill the needs of future application.This thesis gives an in-deep investigation on the quality optimizing and fast 3D-HEVC video coding algorithms, and the major techniques and contributions are listed as follows:1. A quality optimizing algorithm making use of classification and small scale search window is proposed to get the neighboring block disparity vector. This method analyzes the appearance rates of the disparity vector in candidate positions to find the high probability candidate positions that contained the disparity vector. And through these disparity vectors of high probability candidate positions to find the related block in the reference frame, and establish a small scale search window. Then calculate the rate distortion cost between current coding unit and reference block in the small scale search window, modify the primary derivation disparity vector. The rate distortion cost is used to adjust the scale of search window. Experimental results show that our algorithm can remarkably prove the coding quality and with little time lost.2. A quality optimizing algorithm making use of multi-virtual depth maps is proposed to modify the neighboring block disparity vector. This method analyzes the primary obtained neighboring block disparity vector to find the related blocks in reference frame. And through this related block to calculate the virtual depth map of current coding unit. Then analyze the depth value of virtual depth block and neighboring virtual depth blocks, and catalog the max depth value of virtual depth blocks. Through the depth value to calculate the disparity vector to modify the primary neighboring block disparity vector. Experimental results show that our algorithm can remarkably prove the coding quality and with small time lost.3. A fast algorithm making use of group searching and proportion combination is proposed to get the neighboring block disparity vector. This method analyzes the adopting rates of the disparity vector in candidate positions to find the best rate distortion performance candidate positions that contained the disparity vector, and pruning the candidate positions. Through the relevancy of candidate positions that contained the disparity vector divide the remaining candidate positions into several groups. And combine the disparity vector in these groups using proportion factors. In order to get the precise combination factors, use the rate distortion cost to adjust the proportion factors. Through different QP relationship with fix combination factors and adaptive combination factors switch the fix and adaptive combination factors. And analyze the disparity vector to design the fast group searching strategy. Experimental results show that our algorithm can fast get the neighboring block disparity vector and also get the quality improvement of disparity.4. A fast intra mode decision method making use of tile structure and model decision set is investigated for the depth map. In this thesis, take use of the structure similarity and coding model distribution, propose a method using the tile structure to decide the flat area special the single depth area of depth map quickly. Then depend the reference block from current coding unit or sub coding unit of depth map, and take use of the intra model of these reference blocks from depth map or texture map to establish the model decision set for current coding unit. Through this model decision set to early skip the flat intra model. Experimental results show the effectiveness of the method.