基于密度流形上的空间聚类

对于具备空间特性的数据来说,基于密度的聚类方法是一种基本且行之有效的聚类技术.尽管现有很多基于密度的空间聚类算法和技术,但是这些算法多数都假设数据分布于平滑空间.弯曲空间与平滑空间只局部存在相似性.本文的目的在于探讨一种新的基于密度的流形空间聚类,即基于弯曲空间的算法.此算法主要来源于切空间,并适用于非均匀、非线性的数据分布,同时给出了性能分析和实验测试.

Density Based Spatial Clustering on Manifolds

Density based clustering are effective and basic clustering techniques for data with spatial attributes. Although there are many proposed algorithms and applications for density a based spatial clustering, one of its widely used assumptions is that the data is distributed in a smooth space. Manifolds are approximately curved spaces which are locally like smooth spaces but not smooth spaces. The purpose of this paper is to propose new density based clustering algorithms on manifolds, that is, on curved spaces. The newly proposed algorithms will apply for nonlinear, non-uniform data distribution. A simple performance analysis is presented. Experimental data are given for testing the performances of the new algorithms.