Perception-Guided Quality Metric of 3D Point Clouds Using Hybrid Strategy

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Existing full-reference point cloud quality assessment (FR-PCQA) metrics often use unified features, neglecting the human visual system's (HVS) dynamic processing of visual information at different distortion levels.
• The HVS employs distinct strategies for high-quality (distortion detection) and low-quality (appearance perception) samples, a nuance often overlooked by current FR-PCQA methods.

Purpose of the Study:

• To propose a perception-guided hybrid metric (PHM) that bridges the gap in existing FR-PCQA methods by adaptively leveraging HVS visual strategies based on distortion degree.
• To enhance the accuracy of point cloud quality assessment by considering the dynamic nature of human visual perception.

Main Methods:

• PHM employs two adaptive visual strategies: for high-quality samples, it measures visible differences, incorporating the masking effect and texture complexity. For low-quality samples, it utilizes spectral graph theory to assess appearance degradation, analyzing geometric signals and spectral graph wavelet coefficients.
• The metric combines results from these two components using a non-linear approach to generate a final quality score.

Main Results:

• Experiments on five independent databases demonstrate that PHM achieves state-of-the-art (SOTA) performance in point cloud quality assessment.
• PHM offers significant performance improvements across various distortion environments compared to existing metrics.

Conclusions:

• The proposed PHM effectively models the dynamic nature of the HVS for improved FR-PCQA.
• PHM represents a significant advancement in accurately assessing the quality of distorted point clouds, outperforming current SOTA methods.