电子学报 ›› 2021, Vol. 49 ›› Issue (1): 64-71.DOI: 10.12263/DZXB.20191268

• 学术论文 • 上一篇    下一篇

基于卷积神经网络的高效三维模型检索方法

汤磊, 丁博, 何勇军   

  1. 哈尔滨理工大学计算机科学与技术学院, 黑龙江哈尔滨 150080
  • 收稿日期:2019-11-13 修回日期:2020-05-18 出版日期:2021-01-25
    • 通讯作者:
    • 丁博
    • 作者简介:
    • 汤磊 男,1995年出生,山东枣庄人.硕士,主要研究方向为三维模型检索、机器学习.E-mail:tanglei_8520@163.com;何勇军 男,1980年出生,四川南充人.博士,教授,博士生导师,CCF会员.主要研究方向为机器学习、模式识别.E-mail:holywit@163.com
    • 基金资助:
    • 国家自然科学基金面上项目 (No.61673142); 黑龙江省自然科学基金杰出青年项目 (No.JJ2019JQ0013); 黑龙江省普通本科高等学校青年创新人才项目 (No.UNPYSCT-2016034); 哈尔滨市杰出青年人才基金 (No.2017RAYXJ013)

An Efficient 3D Model Retrieval Method Based on Convolutional Neural Network

TANG Lei, DING Bo, HE Yong-jun   

  1. Computer Science and Technology College, Harbin University of Science and Technology, Harbin, Heilongjiang 150080, China
  • Received:2019-11-13 Revised:2020-05-18 Online:2021-01-25 Published:2021-01-25
    • Corresponding author:
    • DING Bo
    • Supported by:
    • National Natural Science Foundation of China (No.61673142); Outstanding Youth Program of Natural Science Foundation of Heilongjiang Province (No.JJ2019JQ0013); Young Innovative Talents Project of Colleges and Universities of Heilongjiang Province (No.UNPYSCT-2016034); Harbin Outstanding Young Talent Fund (No.2017RAYXJ013)

摘要: 目前基于视图的三维模型检索已经成为一个研究热点.该方法首先将三维模型表示为二维视图的集合,然后采用深度学习技术进行分类和检索.但是现有的方法在精度和效率方面都有待提升.本文提出了一种新的三维模型检索方法,该方法包括索引建立和模型检索.在索引建立阶段,选择代表性视图输入到训练好的卷积神经网络(Convolutional Neural Network,CNN)中以提取特征和分类,进而根据特征类别对特征进行组织以建立索引在检索阶段,使用CNN和投票算法将输入模型的代表性视图分类为一个类别,然后仅选择这个类别的特征而不是所有类别的特征进行相似度匹配,因此减少了搜索空间.此外,随着用于检索的视图数量的逐渐增加,一旦可以确定三维模型,检索过程将提前终止.实验的数据选用刚性三维模型数据集ModelNet10,ModelNet40和非刚性三维模型数据集McGill10.结果表明,该方法在提升检索效率的同时,确保检索准确率分别高达94%、92%和100%.

 

关键词: 模型检索, 模型分类, 代表性视图, 卷积神经网络

Abstract: Recently, 3D model retrieval based on views has become a research hotspot. In this method, 3D models are represented as a collection of 2D views, which allows deep learning techniques to be used for 3D model classification and retrieval. However, current methods need improvements on both accuracy and efficiency. We propose a 3D model retrieval method, which includes index building and model retrieval. In the index building stage, representative views are selected and input into a well-learned Convolutional Neural Network (CNN) for feature extraction and classification. Next, the features are organized according to their labels to build indexes. In the retrieval stage, the representative views of the input model are classified into a category with the CNN and voting algorithm, and then only the features of one category rather than all categories are chosen to perform similarity matching. In this way, the searching space for retrieval is reduced. In addition, the number of the used views for retrieval is gradually increased. Once there is enough evidence to determine a 3D model, the retrieval process will be terminated ahead of time. Experiments on the rigid 3D model datasets ModelNet10, ModelNet40, and the non-rigid 3D model dataset McGill10 show that the proposed method can improve the retrieval efficiency substantially while keeping high retrieval accuracy rates at 94%, 92% and 100%, respectively.

 

Key words: model retrieval, model classification, representative views, convolutional neural network

中图分类号: