摘要
为了解决扩展高斯图像在三维模型特征表示上的缺陷,提出了一种多层多分辨率的三维模型特征描述方法.首先用法向主成分分析法对三维模型进行姿态校正,使得三维模型具有平移、缩放、旋转的不变性,然后将三维模型映射到多个扩展高斯球面上,统计各个高斯球面网格上的法向面积分布,并对该分布作球面调和分析得到三维模型的特征描述向量.实验结果表明该方法查全-查准率均好于基于射线的球面映射方法(radialized spheri-cal extent function,REXT),特征描述向量的维数仅为REXT的26.5%.
To solve the major drawback of the extended Gaussian image （EGI） for 3D model representation, we propose a multi-concentric extended Gaussian image with multi-resolution （MCEGI）. Firstly, the 3D model is normalized to the uniform canonical frame to obtain translation, scale and rotation invariance by the normal principal component analysis （NPCA） that we propose. Secondly, a 3D mesh model is decomposed into multi extended Gaussian spheres, and captures its surface area distribution with surface orientation in each Gaussian spherical grid. At last, this distribution function is transformed to spherical harmonic coefficients whose module is regarded as shape descriptors. The experimental results show that the performance of MCEGI on percesion-recall curve is better than REXT,and the dimension of its frature vector is only 26.5% of REXT＇S dimersion.
出处
《武汉大学学报（理学版）》
CAS
CSCD
北大核心
2009年第4期486-490,共5页
Journal of Wuhan University:Natural Science Edition
基金
总装武器装备预研基金资助项目(51306050201)
关键词
扩展高斯图像
法向主成分分析
三维模型检索
extended Gaussian image （EGI）
normal principal component analysis （NPCA）
3D model retrieval
