基于多视角图与卷积神经网络的三维模型检索算法  

3D model retrieval algorithm based on multi-view images and convolutional neural network

作　　者：朱天放 胡书山 余日季[2] 李文靖 ZHU Tianfang;HU Shushan;YU Riji;LI Wenjing(School of Computer Science and Information Engineering, Hubei University, Wuhan 430062, China;School of Art, Hubei University, Wuhan 430062, China)

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062 [2]湖北大学艺术学院,湖北武汉430062

出　　处：《湖北大学学报（自然科学版）》2020年第3期325-333,共9页Journal of Hubei University：Natural Science

基　　金：国家重点研发计划“基于知识驱动的协同创意设计方法及工具”(2016YFB1101702)资助。

摘　　要：随着虚拟现实等领域的迅猛发展,三维模型数量急速增长.针对目前三维模型检索算法存在适用范围小、可拓展性弱、依赖人工干预等不足,提出一种基于多视角图与卷积神经网络的三维模型检索算法.首先,采用多视角图降维描述三维模型,使检索源规整统一;然后,通过卷积神经网络提取视角图高维特征,组成单一的特征描述符集合进行检索;最后,提出一种基于图像熵的多特征组合策略,根据视角图的灰度分布赋予其不同的特征置信度,进一步提高检索精度.经验证,算法最高可达到平均93%的检索精度.为满足快速检索需求,控制检索时间为0.8 s,检索精度高于80%,可有效应用于实际检索系统构建.With the rapid development of virtual reality and other fields,the number of 3D models is increasing sharply.At present,3D model retrieval algorithm has some shortcomings,such as the small scope of application,weak extensibility and dependence on manual intervention.We proposed a 3D model retrieval algorithm based on multi-view images and convolutional neural network(CNN).Firstly,describing the 3D model with the multi-view images to reduce dimension,so the retrieval source was organized and unified.Then,the high-dimensional features of the view map were extracted by the CNN.Finally,we proposed a multi-feature combination strategy based on image entropy,which gave views different weights according to their gray distribution,further improving the retrieval accuracy.Experimental results show that the proposed algorithm can achieve an average retrieval accuracy of 93%.In order to meet the practical requirements,controlling retrieval time to 0.8 s,the accuracy is still higher than 80%,which can be effectively applied in the actual retrieval systems.

关 键 词：三维模型检索 检索算法 深度学习 卷积神经网络 图像熵 多维视角图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]