摘要
Exemplar based clustering aims at assigning samples to the corresponding exemplars that can represent their clusters. Therefore, fast exemplar finding is a very fundamental and significant step in clustering analysis. In this paper, fast exemplar finding based clustering method contains two stages. In the first stage, fast reduced set density estimator (FRSDE) is adopted in order to generate a reduced set of a dataset for guaranteeing the efficiency of exemplar finding. The proposed idea of reduced set based exemplar finding has its basis in two assumptions: (1) the local density of a cluster center is higher than its surrounding neighbors; (2) cluster centers with high local density are at a relatively large distance from any sample with higher local density. In the third stage, samples in the reduced set and remaining set are assigned to different clusters with their corresponding exemplars. We demonstrate the power of the proposed method on synthetic datasets and brain magnetic resonance images (MRI) segmentation.
- 出版日期2016-9
- 单位南通大学