Laplacian多特征映射的三维模型形状分析  

3D Shape Analysis Based on Laplacian Multi-eigenmap

作　　者：韩丽[1,2] 徐建国[1] 黎琳[1] 唐棣[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,大连116081 [2]大连理工大学计算机科学与技术系,大连116024

出　　处：《计算机辅助设计与图形学学报》2015年第11期2142-2148,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61202316);辽宁省高等学校优秀人才支持项目(LJQ2013110)

摘　　要：面向三维模型的统一结构描述与智能检索的技术需求,提出一种Laplacian多特征映射的三维模型形状分析方法.首先提取三维模型的表面形状与体积特征,建立融合测地线距离、角距离和空间体积的多特征相似度矩阵;其次根据Laplacian特征映射算法实现三维模型由空域到谱域的转换以及多谱特征分析;最后通过对Laplacian矩阵特征值之间的本征间隙自适应确定聚类数目,并结合K-means聚类方法实现模型的自动结构识别与分割.实验结果表明,在同一类模型的结构特征提取与统一分割应用中,该方法是高效、鲁棒的,对于实现模型的高层次语义描述、模型配准以及模型检索具有重要的意义.Aiming at the demands of consistent descriptor and intelligent retrieval technology for 3D shapes, we pro-pose a 3D shape analysis based on Laplacian multi-eigenmap. Firstly, we extract the surface and volumetric features of 3D models and construct a multi-feature affinity matrix based on the measurement of geometric distance, angular dis-tance and volumetric distance. Secondly, our method converts 3D spatial domain to spectral domain by using Laplacian multi-eigenmap which effectively reveals the intrinsic invariance and consistent structure among shapes. Finally, we analyze the eigengap to adaptively determine the clustering number and implement automatic structural recognition and segmentation by combining theK-means clustering method. A series of experimental results have shown its robustness and efficiency in shape matching and shape segmentation. Our work has important significance for the high-level se-mantic description, shape registration and shape retrieval.

关 键 词：形状分析 Laplacian特征映射 谱图理论 多特征融合 聚类方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]