A robust fuzzy clustering algorithm for the classification of remote sensing images

A new fuzzy clustering algorithm is presented, that permits one to group data samples even when the number of clusters is not known or when noise is present. The new algorithm is obtained by replacing the probabilistic constraint that memberships across clusters must sum to one with a composite constraint. The composite constraint allows the algorithm to assign low memberships to uncertain data, thus ensuring higher robustness against noise, and avoiding the need to know the number of cluster contained in the data. The results obtained by applying the algorithm to the construction of a land cover map from remote sensed data (LANDSAT) are reported.