A density based cluster extension method
Jian Hou, Chunshi Sha, Lei Chi, Hongxia Cui · 2016
Although numerous clustering algorithms can be found in literature, most of existing algorithms require one or more parameters as input, and their clustering performance usually depends heavily on user-specified parameters. Although some methods have been proposed to determine these parameters automatically, the parameter-tuning problem is still open in general. As a graph-theoretic approach to clustering, the dominant set algorithm uses the pariwise data similarity matrix as input and determines the number of clusters automatically. Although the dominant set algorithm does not require any parameter input explicitly, its clustering results have been found to be influenced by a similarity parameter. In this paper we firstly apply histogram equalization transformation to similarity matrices to remove the dependence on clustering results, and then extend the clusters to obtain better clustering quality. Data clustering experiments validate the effectiveness of our algorithm.