Research On The Fast Intra-frame Coding Algorithm Of Depth Image In 3D-HEVC

With the development of computer and the popularity of smart phones,people's demand for high-definition,ultra-high-definition video and even three-dimensional stereo video become more and more urgent.As a video coding standard,H.264/AVC can't meet the increasing need for video quality.Since depth image can be used to represent the geometric information of the video scene,the 3D-HEVC coding standard introduces depth images to break the relationship between the number of cameras and the baseline of the viewpoint as well as the 3D video quality.Domestic and foreign scholars have actively researched many new coding techniques which have been cited by the 3D-HEVC standard.However,the introduction of these new technologies has dramatically increased the complexity of intra-frame coding in depth images,which seriously hinders the practical application of 3D-HEVC.In this paper,we study the fast intra-frame coding algorithm of depth image and try to reduce the coding complexity and time consuming while ensuring the quality of video.The main research results and innovations of this paper are as follows:A fast DMM1 mode skip algorithm based on corner detection is proposed.The Depth Modelling Mode(DMM)provides a large number of wedge prediction modes(DMM1).Each coding block in depth image performs the cost calculation of all wedge prediction modes,which greatly increases the coding complexity.In this paper,a DMM1 skipping fast algorithm based on corner detection is proposed.The DMM1 prediction mode judgment process is selectively performed by judging whether the current coding block contains corner points: if the coding block has corner points,it performs mode judgment and selection according to the original steps;otherwise,it does not perform DMM1 mode judgment to reduce the coding complexity.A residual mode selection algorithm based on gradient values of texture image horizontal pixels is proposed.The 3D-HEVC introduces a Segment-wise DC Coding(SDC)mode to provide a candidate mode for traditional residual coding based on residual calculation,transform and quantization.However,the introduction of SDC residual coding mode makes it necessary to calculate SDC residual mode cost and non-SDC residual mode cost for all candidate modes in the list of candidate predictive mode.In this paper,the horizontal pixel gradient value of the texture image coding block,which is correspond to the depth image coding block,is selected to obtain the threshold value and evaluate the current depth map coding block.For the depth coding block whose feature value is smaller than the threshold,the non-SDC residual mode judgment is no longer performed and the SDC mode is used to calculate the rate distortion cost for the candidate prediction modes.For the coding block whose texture coding block horizontal pixel gradient value is not less than the threshold value,the prediction mode is sequentially calculated according to the original step.The test of the proposed algorithms uses the video test sequence recommended by Joint Collaborative Team on Video Coding(JVT-3V)and the parameter settings are also set as recommended.Experiments show that the proposed algorithms can reduce the intra-frame coding complexity of the depth image and the coding time while ensuring the quality of synthetic viewpoint video.