Research On The Fast Coding Technology Of Depth Images Oriented To The 3D-HEVC Standard

The quadtree coding partitioning in 3D-HEVC is recursively conducted on different coding units(CU)sizes from 64×64 to 8×8.Besides,the depth mode decision in 3D-HEVC introduces several new coding tools for each CU to improve the coding efficiency.The recursive CU partitioning,together with the depth mode decision at each partition depth level,dramatically increases the computational complexity of depth coding.This thesis is devoted to reducing the complexity of depth coding,as in the following.(1)By taking advantage of the inter-view information and the intermediate RD-cost comparison,an early Merge mode decision for depth maps in dependent views is proposed in this thesis.Experimental results show that the proposed algorithm can provide about 22.1% remarkable time reduction with almost no Bj?ntegaard Delta Bit Rate(BDBR)increased.(2)An early termination method for intra block partitioning is proposed in this thesis,where the first sub-CU selects Depth Intra Skip(DIS)mode as its optimal mode.Since the DIS always has the lowest RD-cost,four times the RD-cost of the first sub-CU is set as the lower bound on one side of the partition decision,and propose a simplified termination decision.Simulation results demonstrate that the proposed algorithm can save about 45.10% of depth coding time with almost no increase in BDBR.(3)An early termination method for intra block partitioning is proposed in this thesis,where the block with a strong consistency of DIS between different partition levels is early coded as the largest CU.The DIS consistency is evaluated by the optimal modes of the current CU and its four sub-CUs based on the new distortion metric at the current partition level.Simulation results demonstrate that the proposed algorithm could provide about 31.04% remarkable time reduction with almost no increase in BDBR.(4)An early determination of depth intra coding is proposed in this thesis,where the coding stage of Intra 2N×2N,Intra N×N or CU Splitting for the current CU can be skipped by the decisions from several learned Decision Trees.Simulation results show that,with the restricted results from the tree leaves by different Gini thresholds,the proposed algorithm could save 41.14%~ 71.63% of the depth coding time with only slight increase in BDBR.