Kalman Filter-Based Coding Optimization And Post-Processing For 3D Point Cloud

A point cloud is a set of points which represent geometric and attribute information.As one of the main representation methods of 3D object or scene,point cloud has been widely used in many fields.With the continuous development of 3D laser scanning technology,the collection accuracy of point cloud is getting higher,and the data volume is also getting larger.Because of the huge data volume of point cloud and the limitation of transmission bandwidth,how to efficiently compress point clouds has become an urgent problem.Motion Picture Experts Group(MPEG)has released a geometry-based point cloud compression(G-PCC)platform which used three attribute coding methods:Predicting Transform(PT),Lifting Transform(LT),and Region Adaptive Hierarchical Transform(RAHT).Among them,the coding efficiency of PT is lower than the latter two.To improve the coding performance of PT,this paper focuses on the attribute encoder optimization and decoder quality enhancement methods.The main research content of this paper includes the following two parts:1.In G-PCC,aiming at improving the coding efficiency of PT,a Kalman filter-based color prediction optimization method is proposed.First,by studying the process of PT and the principle of Kalman filter,a Kalman filter-based prediction optimization method is proposed.Then,the method is further improved by preserving the ground truth color information for some points intermittently.Experimental results show that average-0.5%,-2.8%,-3.6%BD-Rates can be achieved when evaluating the distortion by End-to-End Luma,Cb,Cr,respectively,comparing with the common test platform TMC13v11.0 provided by MPEG.2.To improve the reconstruction quality in decoder,a Kalman filter-based color quality enhancement method is proposed.First,the decoding process of PT is analyzed.Second,as kalman filter performs better on stationary signal,and U(Cb)and V(Cr)color components are more stable than Y(Luma)component,a quality enhancement method special for U and V components is proposed.Experimental results show that average-0.1%,-4.1%and-5.3%BDRates can be achieved when evaluating the distortion by End-to-End Luma,Cb,Cr,respectively,comparing with the TMC13v11.0.