Geometric Clustering of Polsar Data Using the Polar Decomposition
Résumé
This paper presents a new method for geometrical PolSAR clustering based on two fundamental concepts: the polar decomposition of scattering matrices and the Riemannian geometry of their Hermitian factors. The method is applied in a cohesive manner for both coherent and incoherent scatterers. A qualitative comparison is performed with two clustering algorithms based on the covariance framework and two different evaluation metrics: one stochastic – Wishart and one geometric – cosine geodesic. Results on a real dataset show that the final classification better preserves small details and the original texture information in the PolSAR image. In this regard, an improved separation is observed, for example, for certain vegetation fields.
Origine | Fichiers produits par l'(les) auteur(s) |
---|---|
Licence |
Copyright (Tous droits réservés)
|