Separability-based kullback divergence weighting and filter selection for texture classification and segmentation - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2006

Separability-based kullback divergence weighting and filter selection for texture classification and segmentation

Résumé

Featur es computed as statistics (e.g., histo grams) of lo- cal lter responses have been reported as the best descrip- tor s for textur e classication and segmentation. The se- lection of the lter bank remains howe ver a crucial issue , as well as exploiting a rele vant combination of these de- scriptor s. Her e, we propose a novel appr oac h relying on the denition of the textur e-based similarity measur e as a weighted sum of the Kullbac k-Leibler measur es between featur e statistics. The weights are computed accor ding to the maximization of a mar gin based criterion. This weight- ing scheme can also be consider ed as a fast lter selection method: textur e lter response distrib utions are rank ed so that the problem of selecting a subset of lter s reduces to pic king the rst featur es only . Experiments carried out on Brodatz textur es and sonar ima ges show that the proposed weighting method impr oves the classication rates while consider ably reducing the number of the featur es
Fichier non déposé

Dates et versions

hal-02166986 , version 1 (27-06-2019)

Identifiants

  • HAL Id : hal-02166986 , version 1

Citer

Imen Karoui, Ronan Fablet, Jean-Marc Boucher, Jean-Marie Augustin. Separability-based kullback divergence weighting and filter selection for texture classification and segmentation. ICPR'06 : 18th International Conference on Pattern Recognition, Aug 2006, Hong-Kong, Chine. ⟨hal-02166986⟩
23 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More