Full reference point cloud quality assessment using support vector regression

TL;DR

This work tackles full-reference PCQA for 3D point clouds under distortions such as compression, noise, and down-sampling. It introduces FRSVR, which computes five simple FR metrics $S_1$–$S_5$ on a KNN graph derived from the point clouds and then uses a Gaussian-kernel SVR to predict a final quality score, achieving a favorable accuracy-speed balance. The five metrics combine geometric and lightness-information distortions and point-count changes, with formulas $S_1 = \frac{1}{1+E_{p2point}}$, $S_2 = \frac{1}{1+E^{D}_{p2plane}}$, $S_3 = \frac{1}{1+E_{bvar}}$, $S_4 = \frac{1}{1+E_{gvar}}$, and $S_5 = \min\left(1, \frac{|P_D|}{|P_R|}\right)$, mapped to a final score by SVR. Extensive experiments on BASICS, ICIP20, and WPC show FRSVR outperforms conventional FR-PCQA methods in PLCC/SROCC while offering faster computation than complex-feature methods; cross-dataset tests demonstrate reasonable generalization. The approach provides a practical, fast, and accurate FR-PCQA benchmark for guiding point-cloud compression and quality enhancement, with code available at the authors’ GitHub repository.

Abstract

Point clouds are a general format for representing realistic 3D objects in diverse 3D applications. Since point clouds have large data sizes, developing efficient point cloud compression methods is crucial. However, excessive compression leads to various distortions, which deteriorates the point cloud quality perceived by end users. Thus, establishing reliable point cloud quality assessment (PCQA) methods is essential as a benchmark to develop efficient compression methods. This paper presents an accurate full-reference point cloud quality assessment (FR-PCQA) method called full-reference quality assessment using support vector regression (FRSVR) for various types of degradations such as compression distortion, Gaussian noise, and down-sampling. The proposed method demonstrates accurate PCQA by integrating five FR-based metrics covering various types of errors (e.g., considering geometric distortion, color distortion, and point count) using support vector regression (SVR). Moreover, the proposed method achieves a superior trade-off between accuracy and calculation speed because it includes only the calculation of these five simple metrics and SVR, which can perform fast prediction. Experimental results with three types of open datasets show that the proposed method is more accurate than conventional FR-PCQA methods. In addition, the proposed method is faster than state-of-the-art methods that utilize complicated features such as curvature and multi-scale features. Thus, the proposed method provides excellent performance in terms of the accuracy of PCQA and processing speed. Our method is available from https://github.com/STAC-USC/FRSVR-PCQA.

Figures (5)

• Figure 1: The calculation flow of the proposed method. The five scores are utilized to predict the final evaluation score by .
• Figure 2: The sampling process to obtain a sampled point cloud $\mathrm{\ddot{P}_D}$.
• Figure 3: The relationship between the SROCC and the number of neighbors ($K$) for calculating $S_3$ and $S_4$ with the BASICS dataset (validation set).
• Figure 4: The relationship between the average processing time per point cloud [s] and the number of neighbors ($K$) with the BASICS dataset (test set).
• Figure 5: The SROCC and processing time [s] of the proposed method and conventional methods with the BASICS dataset (test set). The upper left of this scatter plot represents a better trade-off between the accuracy and computation speed.