PointPCA: Point Cloud Objective Quality Assessment Using PCA-Based Descriptors
share
With the increasing popularity of extended reality technology and the adoption of depth-enhanced visual data in information exchange and telecommunication systems, point clouds have emerged as a promising 3D imaging modality. Similarly to other types of content representations, visual quality predictors for point cloud data are vital for a wide range of applications, enabling perceptually optimized solutions from acquisition to rendering. Recent standardization activities on point cloud compression have urged the need for objective quality evaluation methods, driving the research community to the development of relevant algorithms. In this work, we complement existing approaches by proposing a new quality metric that compares local shape and appearance measurements between a reference and a distorted point cloud. To this aim, a large set of geometric and textural descriptors is defined, and the prediction accuracy of corresponding statistical features is evaluated in the context of quality assessment. Different combination strategies are examined, providing insights regarding the effectiveness of different metric designs. The performance of the proposed method is validated against subjectively-annotated datasets, showing better performance against state-of-the-art solutions in the majority of cases. A software implementation of the metric is made available here: https://github.com/cwi-dis/pointpca.
READ FULL TEXT
