Co-clustering for hyperspectral images.

Julien Jacques (1, 2) , Cyril Ruckebusch (3)

1 MODAL - MOdel for Data Analysis and Learning
2 ERIC - Entrepôts, Représentation et Ingénierie des Connaissances
3 LASIRE - Laboratoire Avancé de Spectroscopie pour les Intéractions la Réactivité et l'Environnement - UMR 8516

Clustering is often used for hyperspectral images in order to assign sets of pixels into a number of different homogeneous groups called clusters. As a result, pixels in the same cluster have similar spectra, i.e. are close to each other in a certain sense. Clustering is a core technique of the chemometrics toolbox but some limitations can be pointed for hyperspectral imaging. A first limitation of clustering is that it only considers information in the spectral dimension. Another is that it groups whole vectors. This means that if one or a few elements of the vectors differ significantly, the vectors cannot be clustered together. These limitations may result in suboptimal grouping.

Domaines: Traitement des images [eess.IV]; Statistiques [math.ST]

Julien Jacques, Cyril Ruckebusch. Co-clustering for hyperspectral images.. 6th International Conference in Spectral Imaging, Jul 2016, Chamonix, France. ⟨hal-01383918⟩