Research On 3D-HEVC Depth Optimization Based On Coding Information Correlation

Since the multi-view video plus depth(MVD) has good standard compatibility and can be used by depth image-base rendering(DIBR) to synthesize virtual views, it enables the users to choose view-point freely, and is selected as the new generation 3DV standard data format by motion picture expert group(MPEG).High Efficiency Video Coding(HEVC) the newest video coding standard, designed by Video Coding Experts Group(VCEG) and Moving Picture Experts Group(MPEG), has achieved about 50% coding bit-rate saving compared with the existing standard H.264/AVC for equal video quality. 3D-HEVC is the latest video coding standard on the basic of HEVC, especially for coding the MVD information. Compared with video data, depth map has large area of flat region and small amount of sharp edges. Besides, depth map is usually used in synthesizing virtual views rather be watched directly by audience. With the purpose of improving the coding efficiency of depth map, 3D-HEVC adds many efficient coding tools that fit well for the characteristic of depth map based on HEVC. The introduction of coding tools obviously improves the coding efficiency of coding depth, while the coding complexity increases at the same time. Therefore, to decrease the coding complexity of depth map is becoming an essential technology of 3D-HEVC. This paper makes full use of texture-depth coding and depth map intra coding information to give fast coding algorithm of depth map in 3D-HEVC, the main research contents are as follows:By counting and analyzing the optimum coding information in depth map CU and texture CU, we conclude that coding depth of depth map CU is usually smaller than that of its corresponding texture CU. The splitting of CU in different temporal layers(TL) are also researched, finding that coding depth of CU with higher TL tends to be less. The former attributes to depth map’s characteristic having large area of constant value and sharp edges, the latter is because coding depth of CUs with higher TL has more temporal correlation with current CUs. According to these studies, this paper puts forward that LCU’s max coding depth is set as the smaller value between the depth coding of corresponding texture CU and temporal reference CU with higher layer in the process of CU splitting, and that CU splitting is stopped in advance accordingly. The results of experiment show that the proposed algorithm enables the encoding time to reduce by 5.27% compared with the single texture-depth coding algorithm, under the condition that BD-RATE increases by 0.35% on average.Then, this paper studies on the explicit wedgelet coding in depth map coding, especially on the reason of its high complexity. It is widely said the explicit wedgelet coding accounts for over 90% time in the whole intra coding process. After fully analyzing the process of the wedgelet coding, that each PU is predicted by iterating through all the partitioning patterns is found out, from hundreds to thousands, which depends on the size of current PU. On the purpose of solving this difficulty, this paper proposes to connect the best angular direction of HEVC intra prediction and all the partitioning pattern of wedgelet searching mode, which reduces the number of searching partition pattern to about 20% by contrast, and decreases the encoding time of wedgelet coding by 35.76% with the coding BD-RATE increasing 1.34%.