计算机工程 ›› 2018, Vol. 44 ›› Issue (12): 215-221,227.doi: 10.19678/j.issn.1000-3428.0051163

• 人工智能及识别技术 • 上一篇    下一篇

基于多局部显著视图与CNN的三维模型分类

白静1,相潇1,司庆龙1,刘振刚1,秦飞巍2   

  1. 1.北方民族大学 计算机科学与工程学院,银川 750021; 2.杭州电子科技大学 计算机学院,杭州 310018
  • 收稿日期:2018-04-11 出版日期:2018-12-15 发布日期:2018-12-15
  • 作者简介:白静(1982—),女,副教授、博士,主研方向为计算机辅助设计与图形学、机器学习;相潇、司庆龙、刘振刚,硕士研究生;秦飞巍,副教授、博士
  • 基金项目:

    国家自然科学基金(61762003,61502129);浙江省自然科学基金(LQ16F020004);宁夏高等学校一流学科建设项目(NXYLXK 2017A07);国家民委中青年英才计划项目(2016GQR08)

3D Model Classification Based on Multiple Local Salient Views and CNN

BAI Jing 1,XIANG Xiao 1,SI Qinglong 1,LIU Zhengang 1,QIN Feiwei 2   

  1. 1.School of Computer Science and Engineering,Beifang University of Nationalities,Yinchuan 750021,China; 2.School of Computer Science and Technology,Hangzhou Dianzi University,Hangzhou 310018,China
  • Received:2018-04-11 Online:2018-12-15 Published:2018-12-15

摘要:

为提高基于视图的三维模型分类算法准确度,结合多局部显著视图与卷积神经网络(CNN)提出一种新的三维模型分类算法。提取三维模型多视角下的局部视图,引入显著性评价,建立多局部显著视图集合,以合理表征原始三维模型,兼顾数据表示的完整性和多样性。在此基础上,综合单视图CNN,利用bagging策略构建面向三维模型分类任务的集成深度学习模型,从而提高分类器的泛化性和准确率。在ModelNet10数据集上的实验结果表明,该算法可有效提高分类准确率。

关键词: 局部视图, 卷积神经网络, 集成深度学习, 显著视图, 三维模型分类

Abstract:

To improve the accuracy of view based 3D model classification algorithms,a 3D model classification algorithm based on multiple local salient views and Convolutional Neural Network(CNN) is proposed in this paper.Firstly,the local views of a 3D model are extracted from multi perspectives.Then,by evaluating the saliency of each extracted view,multiple local salient views are established to form a reasonable representation of the 3D model,which have the fine properties of integrity and diversity.Finally,integrating the single view based CNN,a 3D model classification oriented ensemble deep learning model is proposed by using the bagging strategy,so as to improve the generalization and accuracy of the classifier.Experimental results on ModelNet10 dataset show that this algorithm can effectively improve classification accuracy.

Key words: local view, Convolutional Neural Network(CNN), ensemble deep learning, salient view, 3D model classification

中图分类号: