Three-dimensional model interest point extraction method and system based on hierarchical learning

Abstract

The invention provides a three-dimensional model interest point extraction method and system based on hierarchical learning. The three-dimensional model interest point extraction method comprises a process of training a layering device for interest points of a three-dimensional model and a detail point extractor , and a process of performing interest point prediction on the three-dimensional modelto be detected, wherein the process of training the layering device of interest points of the three-dimensional model and the detail point extractor comprises the following steps: for any one three-dimensional model in a set D, dividing all manually marked interest points into common interest points and detail interest points; extracting feature descriptors of all vertexes of the surface of the three-dimensional model by utilizing the multi-feature descriptors; assigning values to the interest points and the labels of the points near the interest points by using an activation function so as to train two neural networks respectively; and combining the two trained neural networks through feature vector matching to obtain a layering device and a detail point extractor which can predict interest points of the three-dimensional model. For the three-dimensional model interest point extraction method, the interest point extraction result is obviously superior to that of a traditional algorithm, and the interest point omission rate and repetition rate are low.

Description

technical field [0001] The invention relates to the field of digital geometry processing, in particular to a method and system for extracting interest points of a three-dimensional model based on layered learning. Background technique [0002] The Points of Interest (POIs) of the 3D model, also known as FeaturePoints, are some representative points with geometric and semantic features selected on the surface of the 3D model by imitating the characteristics of human visual perception. 3D model interest points are widely used in 3D model classification and segmentation, mesh deformation editing, face recognition, facial expression recognition and other fields. In recent years, as the application range of 3D models has become wider and wider, the models have become more and more refined, and the improvement of model details has put forward higher requirements for the accuracy and fineness of the 3D model interest point extraction algorithm. [0003] Some interest point extract...

AI Technical Summary

Problems solved by technology

However, since the feature gaps of the interest points in the model details are small, it is difficult to find out these subtle feature gaps from the perspective of the overall learning of the model, so traditional algorithms perform poorly in places where the interest points in the model details are dense

For example, a 3D model of a human body, the traditional algorithm can extract the points of interest corresponding to the hand very well, but it is difficult to accurately extract the points of interest corresponding to each fingertip

[0007] With the continuous advancement of 3D model scanning technology and modeling technology, the grid expression of 3D models has become more and more refined, and the model contains more and more details. Traditional algorithms are no longer competent for finer and finer points of interest. fetch task

Images