Research On Perceptual Quality Assessment Of G-PCC Compressed Point Clou

Point cloud perception quality assessment is the key technology to measure the point cloud perception quality.It can guide the point cloud compression and transmission strategy,so that users can obtain better visual experience.In this paper,the perception quality assessment methods for Geometry-based Point Cloud Compression encoded point cloud are studied in detail.This paper proposes a bitstream-based no reference perceptual quality assessment of G-PCC encoded point cloud,an assessment method of G-PCC encoded rendered point clouds quality,and a machine learning based G-PCC encoded point cloud perception quality assessment method.The main research results are:(1)To monitor the perceived quality of point clouds in real time,a bitstream-based no reference perceptual quality assessment of G-PCC encoded point cloud is presented.Considering the masking effect of texture complexity on human eyes,a method to predict the texture complexity of point clouds by using the information of the bitstream is presented.The quality of G-PCC encoded point cloud is evaluated effectively by combining the texture and geometry coding parameters of G-PCC encoded point cloud bitstream.The SRCC of the model is 0.9447,the PLCC is 0.9465,and the RMSE is 6.8252,indicating that the model has good performance.Compared with the objective index of Graph SIM,the distribution of PLCC and SRCC in this model increased by 0.0223 and 0.0238 respectively,and RMSE decreased by 1.1898.This model only needs to extract the bitstream information in the distorted point cloud,without reference point cloud and fast calculation speed.(2)To evaluate the perceived quality of rendered G-PCC encoded point cloud,an assessment method of G-PCC encoded rendered point clouds quality is presented.Subjective experiments were carried out to study the effect of point size settings on the perceived quality of point cloud rendered.A G-PCC encoded point cloud rendered perception quality evaluation model was established based on the rendering camera distance,the texture and geometry parameters of G-PCC encoded point cloud.The G-PCC encoded point cloud rendering quality was effectively evaluated.Compared with the existing point cloud objective quality evaluation model,this model considers the influence of rendering process on point cloud quality.The SRCC,PLCC and RMSE of this model are 0.9035,0.9198 and 10.3180 respectively,which can be used as a reference for parameter selection in point cloud rendering.(3)In order to improve the performance of G-PCC encoded point cloud perception quality assessment,a machine learning based G-PCC encoded point cloud perception quality assessment model is proposed.Firstly,the geometry and texture characteristics of G-PCC encoded point cloud are extracted,combined with the related characteristics of G-PCC encoded point cloud bitstream layer to support vector regression model and regression random forest training,and point cloud perception quality score is predicted.After debugging and comparing the two models,point cloud perception quality evaluation model established by regression random forest is found to be more effective.Compared with the existing no reference point cloud objective quality evaluation model,this model considers the influence of bitstream layer information and media layer information on point cloud quality and has good performance.Its SRCC is 0.953,PLCC is 0.962,RMSE is 5.929.