Clustering of the Poincare vectors

Data clustering is useful for discovering significant patterns and characteristics in large datasets. In this paper, we address the problem of clustering the Poincare vectors. Three variants of the k-means algorithm and a competitive neural technique are tested and compared. The empirical performance of the different methods in terms of classification and segmentation accuracy is evaluated. The results obtained on real life passive polarimetric images demonstrate the usefulness of such approach for exploiting the polarimetric information.