Research On Coding Algorithm Of Intra Depth Map Based On 3D-HEVC

With the 3D movies, 3D TVs and Auto-stereoscope 3D displays technol-ogy development, 3D video codec technology has become a hot research. In April 2008, the international Moving Picture Experts Group (MPEG) set up the Free-viewpoint Television (FTV) special coding team to enable that the viewers of auto-stereoscope 3D displays can experience stereoscopic vision at different angles and distances. The state-of-the-art video coding standard,HEVC, which is jointly developed by International Telecommunication Union-Telecommunication Standardization Sector (ITU-T) and International Organi-zation for Standardization (ISO), can save about half the bitrate relative to H.264/AVC with the same coding quality. 3D-HEVC, which is 3D extension based on HEVC, has the efficient encoding characteristics of HEVC, and it will be used as the future direction of Three-dimensional Vide (3DV) development.The 3D-HEVC coding framework uses Multiview Video plus Depth (MVD)as the data format of the 3DV coding standard. Compared with the HE VC,the coding side adds the tools for encoding the dependent viewpoint and depth map, in which the Depth-Modeling Mode (DMM) is used as the intra-prediction coding for the depth map. The DMM reduces the bitrate, but it also brings a considerable computational complexity and computational burden.In this paper, the DMM based on 3D-HEVC for intra depth map coding is primarily researched. The original DMM has thousands of candidate modes,which is a disastrous problem in calculating the rate-distortion cost. A lot of excellent scholars have proposed some methods to reduce the number of the candidate models of DMM. In this paper, we propose a fast mode decision al-gorithm based on Hadamard matrix transformation and pre-defined region.First, N of 35 intra prediction modes are inserted into the full-RD search list. Calculate the variance of the prediction coding unit of the depth map and variance as the basis for predicting whether or not the Hadamard matrix trans-formation is performed. Next, calculate the Hadamard matrix transform of the prediction coding unit, which it is required, at its child level. The parental level Hadamard matrix transformation result is obtained by the correspondence be-tween the child level and the parent level. Then, according to the results of the Hadamard matrix transformation, it is possible to know whether the horizontal or vertical edge of the prediction encoding unit is present. If there are verti-cal or horizontal edges in the predication coding unit, the DMM decision will be skipped. Otherwise, the prediction coding unit is discussed in pre-defined region. Determine the area where the split line is located, and then insert its corresponding wedgelet candidate modes into the full-RD search list. Last, the rate-distortion cost in the full-RD search list is calculated to determine the op-timal decision mode.This paper is based on the HTM reference software for the coding time savings and the viewpoint synthesis PSNR calculation experiment. The algo-rithm proposed by Hadamard matrix transformation and regional discussion of the method, skip the DMM or greatly reduce the model of the number of the can-didate models. Compared with the HTM reference software, the coding time is nearly reduced by 58.48%. However, in the aspect of coding quality, the PSNR of synthetic viewpoint synthesis is reduced by 0.3 dB compared with the refer-ence software. It can be concluded that the proposed algorithm greatly reduces the coding complexity and improves the coding efficiency without significant degradation of the coding quality.