Skip to Main content Skip to Navigation
Journal articles

Relaxation d'images de classification et modèles de la physique statistique

Rémi Ronfard 1 Marc Sigelle
1 MOVI - Modeling, localization, recognition and interpretation in computer vision
GRAVIR - IMAG - Graphisme, Vision et Robotique, Inria Grenoble - Rhône-Alpes, CNRS - Centre National de la Recherche Scientifique : FR71
Abstract : We show in this paper the deep relationship between classic models from Statistical Physics and Markovian Random Fields models used in image labelling. We present as an application a markovian relaxation method for enhancement and relaxation of previously classified images . An energy function is defined, which depends only on the labels and on their initial value . The main a priori pixel knowledge results from the confusion matrix of the reference samples used for initial classification . The energy to be minimized includes also terms ensuring simultaneous spatial label regularty, growth of some classes and disparition of some others. The method allows for example to reclassify previous rejection class pixels in their spatial environment . Last we present some results on Remote Sensing multispectral and geological ore images, comparing the performances of Iterated Conditional Modes (ICM) and Simulated Annealing (SA) . Very low CPU time was obtained due to the principle of the method, working on labels instead of gray levels .
Document type :
Journal articles
Complete list of metadatas
Contributor : Rémi Ronfard <>
Submitted on : Wednesday, December 8, 2010 - 2:59:06 PM
Last modification on : Friday, June 26, 2020 - 4:04:03 PM


  • HAL Id : inria-00544599, version 1




Rémi Ronfard, Marc Sigelle. Relaxation d'images de classification et modèles de la physique statistique. Traitement du Signal, Lavoisier, 1992, 9 (6), pp.449-458. ⟨inria-00544599⟩



Record views