Adjunctions on the lattice of dendrograms and hierarchies - Archive ouverte HAL Accéder directement au contenu
Rapport Année : 2011

Adjunctions on the lattice of dendrograms and hierarchies

Fernand Meyer

Résumé

Morphological image processing uses two types of trees. The min-tree represents the relations between the regional minima and the various lakes during flooding. As the level of flooding increases in the various lakes, the flooded domain becomes larger. A second type of tree is used in segmentation and is mainly associated to the watershed transform. The watershed of a topographic surface constitutes a partition of its support. If the relief is flooded, then for increasing levels of floodings, catchment basins merge. The relation of the catchment basins during flooding also obeys a tree structure. We start by an axiomatic definition of each type of tree, min and max tree being governed by a single axiom ; for nested catchment basins, a second axiom is required. There is a one to one correspondance between the trees and an ultrametric half distance, as soon one introduces a total order compatible with the inclusion. Hierarchies obey a complete lattice structure, on which several adjunctions are defined, leading to the construction of morphological filters. Hierarchies are particular useful for interactive image segmentation, as they constitute a compact representation of all contours of the image, structured in a way that interesting contours are easily extracted. The last part extends the classical connections and partial connections to the multiscale case and introduces taxonomies.
Fichier principal
Vignette du fichier
taxonomy.pdf (1.09 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-00605568 , version 1 (02-07-2011)

Identifiants

  • HAL Id : hal-00605568 , version 1

Citer

Fernand Meyer. Adjunctions on the lattice of dendrograms and hierarchies. 2011. ⟨hal-00605568⟩
4790 Consultations
58 Téléchargements

Partager

Gmail Facebook X LinkedIn More