点击次数：

论文类型： 期刊论文

发表时间： 2012-01-01

发表刊物： International Journal of Advancements in Computing Technology

收录刊物： EI、Scopus

卷号： 4

期号： 1

页面范围： 50-58

ISSN号： 20058039

摘要： In recent years, spectral clustering has become one of the most popular modern clustering algorithms. It clusters data points through performing spectral analysis on the associated similarity matrix derived from the data. The particular choice of the similarity matrix is usually defined in a way similar to the Gaussian kernel based on inter-point Euclidean distance in the input space. However the Euclidean distance only reflects the local consistency, spectral clustering gives very poor result on some problems with noise or outliers. In this paper, we develop a novel similarity measure by keeping global smooth consistency. We define the distance between two data points by the minimal distance of the k-neighbors path. We have performed experiments based on both artificial and real data, comparing our method with some other clustering methods. Experimental results show that our method consistently outperforms other clustering methods.