Comparative study of k-means variants for mono-view clustering

摘要

Data clustering analysis is the process of finding similarity between data that are assigned into homogeneous groups and the most heterogeneous as possible among groups. There are several analysis methods in wich K-means clustering algorithm is the widly used in different research areas. Therefore, this paper reviews the most known variants of clustering methods which are K-means, IRP-K-means and FKM. The three main approaches are implemented and tested with a set of image database and using five different generated descriptors with different sizes. An experimental comparative study of the three different clustering methods is presented on basis of purity, accuracy and running time.