Perceptual Quality Assessment of Colored 3D Point Clouds

Honglei Su, Qi Liu, Zhengfang Duanmu, Wentao Liu, Zhou Wang

Abstract

The real-world applications of 3D point clouds have been growing rapidly in recent years, but not much effective work has been dedicated to perceptual quality assessment of colored 3D point clouds. In this work, we first build a large 3D point cloud database for subjective and objective quality assessment of point clouds. We construct 20 high quality, realistic, and omni-directional point clouds of diverse contents. We then apply downsampling, Gaussian noise, and three types of compression algorithms to create 740 distorted point clouds. We carry out a subjective experiment to evaluate the quality of distorted point clouds. Our statistical analysis finds that existing objective point cloud quality assessment (PCQA) models only achieve limited success in predicting subjective quality ratings. We propose a novel objective PCQA model based on the principle of information content weighted structural similarity. Our experimental results show that the proposed model well correlates with subjective opinions and significantly outperforms the existing PCQA models.

Introduction

3D point clouds—collections of points in 3D space defined by geometry and color—are foundational to emerging technologies like VR/AR, autonomous driving, and digital heritage. However, these datasets are massive, often requiring heavy compression that introduces artifacts such as blurring, noise, or geometric distortions. To maintain a high-quality user experience, researchers need reliable Point Cloud Quality Assessment (PCQA) tools. While subjective human rating is the gold standard, it is impractical for real-time applications. This paper introduces the Waterloo Point Cloud (WPC) database and a novel objective metric, IW-SSIMp, designed to mimic human visual perception more accurately than existing 3D metrics.

The Waterloo Point Cloud (WPC) Database

A significant challenge in the field has been the lack of high-resolution, colored 3D datasets. The authors addressed this by constructing the WPC database, which features 20 high-quality source objects ranging from fruits and snacks to tools.

The 20 high-quality source objects used to construct the Waterloo Point Cloud (WPC) database.

To simulate real-world degradation, five types of distortions were applied to these sources: downsampling, Gaussian noise, and three distinct MPEG compression algorithms. This resulted in 740 unique distorted versions. The database's validity was established through a large-scale subjective experiment, collecting 22,800 ratings from 60 human viewers to establish Mean Opinion Scores (MOS).

The IW-SSIMp Model

Existing PCQA models often struggle because they calculate errors directly in 3D space, which is computationally complex and frequently ignores how humans actually perceive objects. The proposed IW-SSIMp model utilizes a projection-based approach:

  1. Multi-view Projection: The 3D object is placed within a virtual icosphere, and 2D "snapshots" are captured from 12 different angles to represent the object's omni-directional appearance.
  2. Information Content Weighting: Recognizing that not all visual data is equally important, the model applies Information Content Weighted Structural Similarity (IW-SSIM). It assigns higher weights to areas with significant visual information, such as sharp edges and complex textures, while ignoring irrelevant background pixels.
  3. Pooling: The scores from these 12 views are averaged to produce a final, robust quality score.
Illustration of the multi-view projection process used in the IW-SSIMp model to convert 3D data into 2D snapshots.

Key Insights and Results

The study revealed that geometry-only metrics are insufficient for colored point clouds; human viewers are highly sensitive to texture artifacts and color bleeding. By focusing on the object itself through information weighting, the IW-SSIMp model mimics human foveal vision and avoids the "background bias" that plagues other projection models.

In performance tests, IW-SSIMp achieved a Pearson Linear Correlation Coefficient (PLCC) of 0.8504 and a Spearman Rank Correlation Coefficient (SRCC) of 0.8481 on the WPC database. It significantly outperformed 13 state-of-the-art models, including standard MPEG metrics. Furthermore, cross-database validation on benchmarks like SJTU-PCQA and M-PCCD proved the model's stability and generalizability.

Conclusion and Future Work

While IW-SSIMp represents a major step forward, challenges remain. The computational cost of high-resolution projections is high, suggesting a need for more efficient viewpoint selection. Additionally, while this study used passive video viewing, future research should explore interactive environments where users freely rotate objects. Finally, the development of "No-Reference" models—which can assess quality without the original source—remains a critical goal for the field.

Summary updated

Abstract

The real-world applications of 3D point clouds have been growing rapidly in recent years, but not much effective work has been dedicated to perceptual quality assessment of colored 3D point clouds. In this work, we first build a large 3D point cloud database for subjective and objective quality assessment of point clouds. We construct 20 high quality, realistic, and omni-directional point clouds of diverse contents. We then apply downsampling, Gaussian noise, and three types of compression algorithms to create 740 distorted point clouds. We carry out a subjective experiment to evaluate the quality of distorted point clouds. Our statistical analysis finds that existing objective point cloud quality assessment (PCQA) models only achieve limited success in predicting subjective quality ratings. We propose a novel objective PCQA model based on the principle of information content weighted structural similarity. Our experimental results show that the proposed model well correlates with subjective opinions and significantly outperforms the existing PCQA models. The database has been made publicly available to facilitate reproducible research at https://github.com/qdushl/Waterloo-Point-Cloud-Database.
BibTeX
@article{su2023perceptual,
  author = {Honglei Su and Qi Liu and Zhengfang Duanmu and Wentao Liu and Zhou Wang},
  title = {{Perceptual Quality Assessment of Colored 3D Point Clouds}},
  journal = {IEEE Trans. Vis. Comput. Graph.},
  year = {2023},
  url = {https://openreview.net/forum?id=muxzo22ptQ},
}
Figure