With the pairwise data similarity matrix as input, dominant sets clustering has been shown to be a promising clustering approach with some nice properties. However, its clustering results are found to be influenced by the similarity parameter used in building the similarity matrix. While histogram equalization transformation of the similarity matrices removes this influence effectively, this transformation causes over-segmentation in the clustering results. In this paper we present a density based cluster extension algorithm to solve the over-segmentation problem. Specifically, we determine the density threshold based on the minimum possible density inside the dominant sets and then add new members into clusters if the density requirement is satisfied. Our algorithm is shown to perform better than the original dominant sets algorithm and also some other state-of-the-art clustering algorithms in data clustering and image segmentation experiments.
CITATION STYLE
Hou, J., Sha, C., Xu, E., Xia, Q., & Qi, N. (2015). Density based cluster extension and dominant sets clustering. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9069, pp. 262–271). Springer Verlag. https://doi.org/10.1007/978-3-319-18224-7_26
Mendeley helps you to discover research relevant for your work.