Revisit Point Cloud Quality Assessment: Current Advances and a Multiscale-Inspired Approach

The demand for full-reference point cloud quality assessment (PCQA) has extended across various point cloud services. Unlike image quality assessment where the reference and distorted images are naturally aligned in coordinate and thus allow the point-to-point (P2P) color assessment, the coordinates and attribute of a 3D point cloud may both suffer from distortion, making the P2P evaluation unsuitable. To address this, PCQA methods usually define a set of key points and construct a neighborhood around each key point for neighbor-to-neighbor (N2N) computation on geometry and attribute. However, state-of-the-art PCQA methods often exhibit limitations in certain scenarios due to insufficient consideration of key points and neighborhoods. To overcome these challenges, this paper proposes PQI, a simple yet efficient metric to index point cloud quality. PQI suggests using scale-wise key points to uniformly capture distortions within a point cloud, along with a mild neighborhood size associated with each key point for compromised N2N computation. To achieve this, PQI employs a multiscale framework to obtain key points, ensuring comprehensive feature representation and distortion detection throughout the entire point cloud. Such a multiscale method merges every eight points into one in the downscaling processing, implicitly embedding neighborhood information into a single point and thereby eliminating the need for an explicitly large neighborhood. Further, within each neighborhood, simple features like geometry Euclidean distance difference and attribute value difference are extracted. Feature similarity is then calculated between the reference and distorted samples at each scale and linearly weighted to generate the final PQI score. Extensive experiments demonstrate the superiority of PQI, consistently achieving state-of-the-art performance across several widely recognized PCQA datasets. Moreover, PQI is highly appealing for practical applications due to its low complexity and flexible scale options

The framework of proposed multiscale structural similarity-based PQI which consists of four stages: multiscale downsampling (DS), neighborhood construction, feature aggregation, and similarity calculation. The red points are key points from the reference point cloud.

We compared the currently well-known full-reference point cloud quality evaluation methods, including: p2p0-MSE, p2pl-MSE, p2po-PSNR, p2pl-PSNR, Y PSNR, YUV PSNR, PointSSIM, PCQM, GraphSIM, MS-GraphSIM and MPED in the SJTU-PCQA dataset.

This paper proposes a point cloud quality assessment (PCQA) method named PQI. We first reveal that the conventional point-to-point assessment is no longer adequate for PCQA due to the mismatch between the reference and the distorted point clouds and suggest a neighbor-to-neighbor (N2N) manner instead. Further, we revisit and analyze state-of-the-art methods, showing their limitations and unreliability on key point selection, neighborhood construction, etc. To overcome these limitations, PQI suggests the utilization of a multiscale mechanism to generate key points of varying scales, ensuring a comprehensive perception of various distortions in a point cloud. Additionally, PQI employs geometry Euclidean distance and attribute magnitude difference to characterize a point cloud, making it easy to implement. Finally, the feature similarities of different scales are compared and linearly combined for the final scoring. Experimental results confirm the remarkable performance of PQI, showcasing its significant superiority over existing approaches across publicly recognized datasets.
Currently, we average the PQI scores of different scales to predict the final score. Additionally, we apply the same size of local neighborhood field (i.e., N = 10) across all scales for unified processing. In future work, we will continue to exploit the utilization of varying local neighborhood sizes and adjust weighting parameters for different scales, with the aim of further improving overall performance.