New methods for cluster selection in unsupervised fuzzy clustering

Cluster analysis has been playing an important role in solving many problems in pattern recognition and image processing. The fuzzy clustering has been widely used in pattern recognition to search for substructures in a multidimensional data space. Unsupervised clustering algorithms have a variable number of clusters as opposed to supervised clustering algorithms. Unsupervised clustering algorithms utilize various criteria to decide if and how to introduce a new cluster center. Three new methods for selection of a new cluster center in the K-means fuzzy clustering algorithm are presented in this paper. The comparison of new techniques is done with respect to cluster validity and speed of convergence. The technique is applied to the problem of segmentation of human head images obtained by Computed Tomography (CT). Experiments have been performed to compare the proposed techniques with respect to convergence speed and cluster validity measures.