| The development of computer technology has provided convenience to people’s lives.As a kind of important multimedia information types,3D model has provided many benefits to the fields such as games and medical treatment in its rich form.Content-based 3D model retrieval technology is designed to make it easy for people to get the 3D model they want,which is designed to help people search 3D models similar to the target 3D model.Non-rigid 3D model will put forward higher requirements for retrieval algorithms.In this paper,the main work of non-rigid 3D model is as follows: 3D model retrieval method is divided into two types based on the model itself and based on the summary view.This paper summarizes the statistical learning methods commonly used in the field of 3D model retrieval,including bag-of-features method and dimensionality reduction method.A method based on geodesic distance and MDS is proposed to generate canonical forms and compare it with similar methods.A local feature is constructed on the point cloud.The main method is to construct a unique coordinate system for the defined local area and to calculate the vertex distribution in the local area.And then classified using bag-of-features method after dimensional reduction.Then the histogram distribution is calculated for the 3D model.Finally,the similarity between the models is obtained by comparing the histogram distribution.The similarity ranking results are the retrieval results.As a contrast,another method based on local feature extraction is implemented,which transforms the 3D mesh into oriented point set with position and direction information,then constructs local features by calculating the position angle relation among the vertices in the defined local areas.Finally,we compared the two method above.Through several experiments,the 3D model retrieval algorithm based on local feature and statistical learning has better effect on the general data set.For the non-rigid 3D model retrieval,the canonical-form method is an effective preprocessing method;the retrieval method based on local feature and statistical learning is a kind of effective 3D model retrieval framework. |
