Skip to Main content Skip to Navigation

Traitement d'image à voisinages adaptatifs généraux.

Abstract : This thesis deals with multiscale and adaptive (both with spatial and intensity settings) processing of gray tone images. From a punctual characterization, an image is represented with a set of local neighborhoods, called general adaptive neighborhoods (GANs). For each point of the image to be studied, an increasing collection of GANs is defined, allowing a context-dependent multiscale analysis to be performed. These GANs are adaptive in the sense that each neighborhood spatially coincides with the local structure of the seed point, following the radiometric, morphological, geometrical or textural characteristics to be analyzed. In addition, the GANs are physically relevant since they depend on the physical and/or psychophysical nature of the image to be studied. Consequently, the GANs are adapted to linear images or imaging systems, but also nonlinear and/or bounded range images such as transmitted light images, practical digital images or the human brightness perception system. This image representation, based on general adaptive neighborhoods, allows efficient image processing tools to be built. These GANs naturally constitute operational windows for local image transformations. Firstly, adaptive mathematical morphology is introduced using GANs as (adaptive) structuring elements. The resulting transformations satisfy the usual properties of classical morphological operators. Moreover, in several important and practical cases, the adaptive morphological operators are connected, that is of great topological importance, in comparison to the usual ones which fail to this property. Secondly, Choquet filtering is extended with the GANs, generalizing several nonlinear operators such as rank-order filters. Beyond, the GANs enable to introduce local adaptive descriptors for gray tone images, such as orientation or thickness. These local measures lead on the definition of new GANs allowing a more significant spatial analysis to be performed, or the solving of practical applications in image processing and analysis. In this thesis, the general adaptive neighborhood image processing (GANIP) is applied in the fields of image restoration, image enhancement or image segmentation. This approach promises large theoretical prospects and should permit the devising of several image processes responding to concrete problems.
Complete list of metadata

Cited literature [219 references]  Display  Hide  Download
Contributor : Andrée-Aimée Toucas <>
Submitted on : Thursday, November 8, 2012 - 2:59:27 PM
Last modification on : Thursday, October 22, 2020 - 11:34:03 AM
Long-term archiving on: : Saturday, February 9, 2013 - 3:47:57 AM


  • HAL Id : tel-00749859, version 1


Johan Debayle. Traitement d'image à voisinages adaptatifs généraux.. Traitement du signal et de l'image [eess.SP]. Ecole Nationale Supérieure des Mines de Saint-Etienne, 2005. Français. ⟨NNT : 383IVS⟩. ⟨tel-00749859⟩



Record views


Files downloads