Centroid averaging algorithm for a clustering ensemble

A collective approach to cluster analysis is considered in the paper. An algorithm of centroid averaging is proposed. The algorithm allows constructing the consensus partition of a dataset into clusters, using a set of partitions built with any centroid-based algorithm. We discuss results of applying the proposed algorithm to modeled data and for the segmentation of hyperspectral images with noise channels. Some details of implementation in a multithreaded environment that allows increasing the algorithm performance are given.