基于码流的G-PCC压缩点云无参考感知质量评估

为了实现对采用G-PCC编码的点云质量的实时监控，提出一种基于码流的无参考点云感知质量评估模型。首先根据主观实验的结果分析确定几何无损时点云的感知质量与纹理量化参数之间的关系，然后使用纹理量化参数和纹理比特率来预测纹理复杂度，结合空域掩盖效应建立几何无损时的点云质量评估模型。然后研究位置量化尺度对点云下采样质量的影响，发现纹理量化参数与位置量化尺度对点云质量的影响相互独立，并最终得到完整的点云质量评估模型。对该模型在WPC4点云数据库中进行测试，其SRCC为0.9447，PLCC为0.9465，RMSE为6.8252，表明该模型有良好的性能，与现有性能最好的GraphSIM客观指标相比，该模型指标的PLCC和SRCC分布分别提高了0.0223和0.0238，RMSE降低了1.1898。 

Bitstream-based no Reference Perceptual Quality Assessment of G-PCC Encoded Point Cloud

? ?In order to realize the real-time monitoring of point cloud quality， this paper proposed a bitstream-based no reference model for perceptual quality assessment of G-PCC encoded point cloud. Firstly， according to the analysis of subjective experimental results， the relationship between perceived quality of point cloud and texture quantization parameters was determined when geometry coding was lossless. Then texture complexity was estimated using texture quantization parameter and texture bitrate. The point cloud quality assessment model was established according to spatial masking effect when geometry coding was lossless. The influence of location quantization scale on sub-sampling quality of point cloud was studied. It was found that the influence of texture quantization parameter and location quantization scale on point cloud quality were independent of each other. Finally， a complete evaluation model of point cloud quality was obtained. The model was tested on WPC4 point cloud database. The SRCC， PLCC and RMSE of the model are 0.9447， 0.9465 and 6.8252， respectively， in the WPC4 point cloud database， indicating that the model has good performance. Compared with the existing objective indexes of GraphSIM with the best performance， the PLCC and SRCC distribution of this model index increased by 0.0223 and 0.0238， and RMSE decreased by 1.1898.