Robust clustering based on global data distribution and local connectivity matrix

A new method of clustering analysis, which is based on integration of graph theoretical method and fuzzy objective function algorithm, is developed. The connectivity matrix derived from fuzzy limited neighborhood graph and the measurement for similarity and dissimilarity are utilized to build a new fuzzy objective function that unifies global data distribution and local spatial information. In some sense, both the traditional graph theoretical method and objective function algorithm are special cases of our algorithm.