Clustering with a mixture of self-organizing maps

In clustering, usually a single point or vector is used as a cluster prototype. The idea of clustering about principal curves has been recently introduced. Principal curves are functions that can characterize a set of nonlinear data. One way to create a principal curve is to apply a one-dimensional self-organizing map to the multidimensional data. In this paper, the mixture of self-organizing maps algorithm is presented. Results with respect to a color image segmentation task are shown and discussed.