Detecting 3D Feature Points Via Hierarchical Learning Strategy

With the continuous improvement of computer hardware performance and the rapid development of computer graphics algorithms,three-dimensional models have become more and more widely used in medical,animation,architecture,games and other fields,and people’s research on three-dimensional models has become more and more in-depth.As the most basic element reflecting the geometric and semantic features of the 3D model,the feature points of the 3D model are widely used in the fields of 3D model viewpoint selection,classification and segmentation,shape retrieval,face recognition,etc.Therefore,the detection of 3D model feature points has always been research hotspots and focuses in the field of computer graphics.In recent years,with the increasing accuracy of 3D model scanning technology and the continuous improvement of modeling technology,the model has become more sophisticated,and the detailed information contained in the model itself has become more and more abundant.The traditional feature point detection algorithm is no longer capable of accuracy Increasing feature point detection tasks.To this end,this paper proposes a three-dimensional mesh model feature point detection algorithm based on hierarchical learning.The algorithm flow is as follows:First extract the feature vectors of all vertices on the surface of the 3D mesh model;then divide the manually labeled feature points into sparse feature points and dense feature points.For sparse feature points,you can train a neural network on the entire 3D mesh model.Feature points,neural networks can be trained separately in areas where feature points are densely distributed;then feature vector matching operations are performed on two trained neural networks to obtain a classifier,which is used to extract and predict feature points of the three-dimensional model.During the test,a new three-dimensional model is input,after extracting the feature vectors of all vertices on the surface of the model,the extracted feature vectors are input into the trained classifier for prediction operations,and then the features are extracted using an improved density peak clustering algorithm point.The algorithm of this paper adopts a layered learning strategy,which solves the problem that the accuracy of the feature point detection of the traditional algorithm at the details of the 3D model is not high enough.The 20 types of models in the public data set SHREC’11 are used as the evaluation criteria,and compared with three traditional 3D model feature point detection algorithms.The experimental results show that the proposed algorithm has higher accuracy in extracting feature points and generates missing feature points.There are fewer false feature points,which is of great help to solve the problem of detecting feature points of 3D models with increasing fineness.