Point Clustering via Voting Maximization

Abstract

In this paper, we propose an unsupervised point clustering framework. The goal is to cluster N given points into K clusters, so that similarities between objects in the same group are high while the similarities between objects in different groups are low. The point similarity is defined by a voting measure that takes into account the point distances. Using the voting formulation, the problem of clustering is reduced to the maximization of the sum of votes between the points of the same cluster. We have shown that the resulting clustering based on voting maximization has advantages concerning the cluster's compactness, working well for clusters of different densities and/or sizes. In addition, the proposed scheme is able to detect outliers. Experimental results and comparisons to existing methods on real and synthetic datasets demonstrate the high performance and robustness of the proposed scheme.