CE Re-clustering: Reconstruction of Clustering Results by Three-Way Strategy

摘要

Many existing approaches to clustering are based on a two-way strategy that does not adequately show the fact that a cluster may not have a well-defined boundary. In this paper, we propose a Contraction and Expansion Re-clustering (C&E Re-clustering for short) model based on three-way strategy. The model utilizes the ideas of erosion and dilation in mathematical morphology. Contraction is used to shrink clusters while expansion is used to stretch clusters obtained by using an existing clustering algorithm. The difference between the results of contraction and expansion is regarded as the fringe region of the specific cluster. Therefore, a three-way explanation of the cluster is naturally formed. A C&E NJW re-clustering algorithm is proposed and the results on synthetic data set show that such a strategy is effective in improving the structure of clustering results.