基于骨架特征点的三维模型聚类  

CLUSTERING 3D MODEL BASED ON SKELETON CHARACTERISTIC POINTS

作　　者：邱琳[1] 陈晓云[1] 

机构地区：[1]福州大学数学与计算机科学学院,福建福州350108

出　　处：《计算机应用与软件》2012年第6期162-166,共5页Computer Applications and Software

基　　金：福建省高等学校新世纪优秀人才支持计划基金项目(XSJRC2007-11)

摘　　要：针对三维模型聚类问题,提出一种基于骨架特征点的三维模型聚类算法。该算法首先对三维模型的二维投影图进行预处理,然后对投影图进行二级分解,提取小波分解后投影图的骨架特征点,并采用质心距离将其序列化。针对骨架特征序列非等长问题,采用基于DTW度量的K-medoids聚类算法进行聚类。最后在PSB数据集上进行实验,结果表明,该方法能够得到较好的聚类效果,对处于各个姿态的三维模型进行有效地聚类。In this paper,a new algorithm is proposed for clustering 3D model based on skeleton characteristic points for 3D model clustering issue.In the algorithm,firstly the two-dimensional projection of three-dimensional model is pre-processed,then second-level decomposition is performed on projection image to extract the skeleton characteristic points of the projection image decomposed by wavelet,and serialising them with centroid distance method.To solve the problem of unequal length in skeleton feature sequence,K-medoids clustering algorithm based on DTW metric is used for clustering.Finally,experiments are conducted on PSB data set,results indicate that the method can get better clustering accuracy,the clustering is effective on 3D models in various gestures.

关 键 词：三维模型 骨架 DTW动态时间弯曲 k中心聚类 小波分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]