Study On Fast Optimization Algorithms For Depth Map Coding In 3D-HEVC

With the rapid development of multimedia communications technology and processing capabilities of various devices.3D video becomes more and more popular in our life.Since the multi-view video coding standards based on H.264 can not meet the compression of the increasing 3D video data,the Joint Collaborative Team on 3D Video Coding Extension Development(JCT-3V)makes the new multi-view video coding standards,3D-High Efficiency Video Coding.Although 3D-HEVC achieves high coding efficiency,it results in a high computational complexity which seriously hinders its practical application.Therefore,how to reduce the computational complexity of 3D-HEVC as much as possible while keeping the coding efficiency is becoming a hot point in the field of video technology.At present,the commonly used 3D video format is Multi-view Video Plus Depth,in which the depth map plays an important role in virtual view synthesis.However,the depth map coding also results in huge computational complexity.Therefore focusing on fast optimization algorithms for the depth map coding in 3D-HEVC,this thesis deeply analyses the characteristics of depth map,with the goal of reducing the computational complexity while keeping the coding efficiency.The major contribution of this paper can be described as follows:Firstly,a low complexity depth intra coding for depth map coding in 3D-HEVC is proposed based on depth classification.According to the feature of depth map,the algorithm divides the intra modes into three classes: smooth,direction and DMM.The HOG is used to extract the feature of depth prediction unit,and then the extracted feature is judged by the SVM trainer to select the corresponding classes.The selected prediction modes are further checked by RDO to choose the optimal mode.The results in this paper show that the proposed algorithm can reduce the encoding time by 34.85% on average,while the BD-Rate reduces the 0.14%.Secondly,a method for accelerating segmentation of depth block CU is proposed.A recursive method of segmenting CU in the 3D-HEVC consumes a great deal of computational load.Therefore,this thesis presents a fast method to terminate CUrecursive as follows.The sum of the absolute values of the variance and the diagonal pixel difference are calculated firstly,and then compared with the pre-set threshold to early terminate the CU division.Shown as the experimental results,the proposed algorithm in this thesis reduces computational complexity by 9.73% on average,while the BD-Rate is only increased by 0.02% compared with the original3D-HEVC.Finally,this thesis developed a skip SDC coding scheme.It is observed from the experimental statistics that the usage of SDC coding is closely related to the smoothness of the current PU.If the PU is relatively smooth,the possibility of selecting SDC coding is high.Therefore,after establishing the full search list,the algorithm calculates the sum of the absolute values of the pixels in outer circle of the current PU,and then compared with the pre-set threshold to determine whether to skip the non-SDC coding so that the computational complexity is reduced.Shown as the experimental results,the proposed algorithm in this thesis can reduce the encoding time by 10.64% on average compared with the 3D-HEVC,while BD-Rate is only increase by 0.16%.In addition,this paper integrates all the three optimization algorithms to optimize the intra coding process of depth map.Shown as the experimental results the proposed algorithm can reduce the coding time by 43.09%on average compared with the original 3D-HEVC,while the BDBR is only increased by 1.06%.In summary,this thesis proposed a set of fast optimization algorithms for depth map coding in 3D-HEVC,which are able to effectively reduce the computational complexity while maintaining the 3D-HEVC coding efficiency.This thesis is of a certain theoretical and practical value to promote the application of 3D-HEVC.