A Study On Efficient Compression Of 3D Color Point Cloud And Quality Evaluation

With the development of the technologies,the requirements are gradually improved for the point cloud of real scene from 3D scanner;the high accuracy and rich color information are the main characters of the 3D data,and the data is widely used in modeling,simulation and roaming.For the vast amount of color point cloud data,some challenges must be faced.The compression is the elementary question,how to find effective ways to compress is a problem that needs to be solved;at the same time,the distortion of point cloud exists after transmission or compression,so the quality evaluation must be given to evaluate the loss of the quality.So the research on compression and quality evaluation of the color point cloud will have theoretical and practical significances.Combining the times,the data we studied has two characters: firstly,our data is mainly focuses on color point-based data,it can be processed directly without format conversion,this can reduce the distortion;Secondly,our data is color point cloud,the point cloud contains position information,and also contains color information,which can express the objective world more accurately;so,the paper focuses on color point cloud.This paper mainly discusses the compression and the quality evaluation of the color point cloud,including the removal of the outliers of point cloud,clustering segmentation based on 6D characteristics,then we use the clustering results and the DCT transform to compress the 3D point cloud;For the quality evaluation,we discuss the objective evaluation according to the subjective evaluation,and the main content includes:For the outlier noise of the scanned point cloud,a 3D statistical filtering algorithm is adopted;the algorithm can kick off the outliers that deviate from the specified threshold.The experiment shows that the given 3D statistical filter can effectively filter the outliers in point cloud,and the algorithm has better universality.A clustering segmentation algorithm based on 6D features is proposed for the color point cloud.A clustering segmentation must select proper characteristics according to the corresponding applications,in our paper we focus on the compression,so there are two characteristics we chosen: one is the space(xyz)distribution of point cloud,a Mean Shift algorithm based on Gauss Kernel function is adopted,this can guarantee rapid convergence to the density center of the cluster;the other characteristic is the chromaticity space(rgb),the color information can guarantee the maximized consistency in data region.The three dimensional space coordinates and three dimensional chromaticity Spaces are used in our clustering segmentation,so we call them 6D characteristics.The experimental results show that this algorithm has better segmentation results,for it considers both the density and the color information.In this paper,we project the clustered point cloud to the implicit plane,and propose an efficient compression algorithm based on DCT transform in the implicit plane.For the color point cloud constituted with pure points,firstly we fit the plane equation using the RANSAC algorithm for a clustered segmentation,and project the points to the implicit plane;secondly,the best bounding box is calculated for the projected points,then the bounding box is regulated by 4 by 4 grids;finally,2D DCT transformation and entropy coding are performed for the regular point cloud,thus the color point cloud is compressed.This compression algorithm not only realizes the high efficient compression for the color point cloud,but also the compression is fast.The compression quality is quite better.A SCSIM(Structural and Color Similarity)objective evaluation algorithm is presented for the color point cloud.At present,there is no uniform standard for the subjective and objective evaluation of point cloud,this paper refers to the subjective evaluation and SSIM(Structural Similarity),a SCSIM algorithm is proposed to evaluate the quality of the color point cloud.Firstly,we calculate the weight of geometry saliency in the structured clustering segmentation of the point cloud,and get the differences of weighted histogram between the original point cloud and degraded point cloud,give out the structural similarity evaluation;secondly,the color similarity is given out based on the color histogram in the structural clustering segmentation;finally the objective quality evaluation results are given based on the fusion of structure and color similarity.The experimental results show that the objective evaluation and subjective evaluation match well.