摘要
三维形状识别是近年来较为热门的研究方向,针对其中的三维模型形状的表达方法和识别问题,提出一种多分支卷积神经网络下的三维模型识别方法.该方法通过对三维模型进行球面深度投影得到球面全景图;为了提高识别精度,将每个模型的球面全景图从多个角度展开,创建多幅平面图像作为识别系统的输入;识别系统使用多分支的卷积神经网络,并将多幅全景图进行整合分析,最终得到一个三维模型的识别结果.对三维模型进行分类和检索的实验结果表明,文中方法的识别效果优于近年来的前沿方法,对三维模型进行检索的准确度甚至超过了多视图识别方法.
3D shape recognition is a hot topic in recent years.This paper proposed a3D model recognition method with multi-branch convolutional neural network(CNN)to address the problems of3D shape representation and recognition.The inputs of the proposed method are spherical panoramas by deep spherical projection of3D models;to improve recognition accuracy,the spherical panorama of the shape first unfolded on various orientations to produce multiple rectified images as input of recognition frame;the recognition system consists of a multi-branch CNN,which analyzes the panoramas as a whole to produce the final recognition result.The experiment results of retrieval and classification on various of3D dataset showed that the performance of our method is better than the state-of-the-art methods,and the retrieval accuracy outperforms that of multi-view method.
作者
冯元力
夏梦
季鹏磊
周潇
曾鸣
刘新国
Feng Yuanli;Xia Meng;Ji Penglei;Zhou Xiao;Zeng Ming;Liu Xinguo(State Key Laboratory of CAD&CG, Zhejiang University, Hangzhou 310058;Software School of Xiamen University, Xiamen 361005)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2017年第9期1689-1695,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61379068)
关键词
卷积神经网络
形状识别
球面投影
全景图
convolutional neural network
shape recognition
spherical projection
panorama