Extraction of Points of Interest on 3D Models Based on Voting and Dynamic Grouping

Extracting points of interest on 3D model is a basic problem in computer graphics, aiming at this problem, a method using voting mechanism and dynamic grouping is proposed, which mainly includes 3 modules. Among them, the vertex encoder module fuses neighborhood and global information of vertices to generate vertex semantic sequences; The probability distribution prediction network module maps the vertex semantic sequence into a probability voting sequence and a confidence sequence to generate a probability distribution of points of interest; The dynamic grouping module groups and extracts points of interest from the probability distribution by setting different probability thresholds. The proposed method is tested on SHREC2011 and KeyPointNet data sets and use FNE, FPE, BHD and CD as evaluation metrics. Compared with traditional methods, FNE is reduced by at least 0.2 and FPE is reduced by at least 0.18; compared with existing machine learning methods, BHD is reduced by 0.011 on average and CD is reduced by 0.002 on average, the proposed method has a significant improvement for extracting points of interest.