Performance of Point Cloud- and Image-Based Quality Metrics on Geometry-Only Point Clouds and Textured Point Clouds

This paper presents an analysis of the performance of point cloud-based and image-based quality metrics on dynamic point clouds, based on subjective quality assessment of their geometry-only and textured versions. The subjective quality assessment was conducted as an expert viewing with 2D video representations of the point clouds using pre-defined camera paths for rendering. It is considered to be a first-time experiment assessing geometry-only rendering with dynamic point-clouds. The study includes correlation measures and a statistical analysis on the reliability of objective metrics. The suitability of the metrics for the evaluation task and the subjective assessment of the geometry-only case are discussed. The results indicate that geometry-based PSNRY as well as image-based metrics can provide reliable results for textured point clouds. Based on the subjective results, the objective assessment of geometry-only point clouds is found to be less reliable than for the textured ones. Other findings report a divergence between results of the geometry-based D1 and D2 metrics and the subjective impact of geometry errors.