Blind Point Cloud Quality Assessment via 3D Visual Saliency and Point-Based Neural Network

No-Reference/Blind Point Cloud Quality Assessment (NR-PCQA) aims to evaluate the quality of a point cloud in a manner similar to human judgment, without needing a reference. In the literature, most NR-PCQA methods do not accurately incorporate human visual information, which is crucial for quality evaluation. In this paper, we present a novel point-based NR-PCQA method that integrates 3D visual saliency into the quality assessment process. First, we generate a 3D saliency map, which is then used to weight the corresponding point cloud. After that, the salient point cloud is fed into a 3D point-based encoder to estimate the objective quality score. Extensive experiments conducted on two publicly available PCQA databases demon-strate that our proposed model surpasses existing state-of-the-art methods.