Skip to Main content Skip to Navigation

Mise en oeuvre d'un système adaptatif de segmentation d'images

Christophe Rosenberger 1
1 Monétique & Biométrie
LASTI - Laboratoire d'Analyse des Systèmes de Traitement de l'Information
Abstract : Image processing reveals an increasing interest as image becomes a support as a privileged source of informations. The quality of interpretation of an image considerably depends on its segmentation result. Despite of the diversity of methods, segmentation results still remain unsatisfied and vary a lot a function of a given technique. In order to contribute to solve this problem, an adaptive image segmentation system is proposed.

After a bibliographic study providing the enumeration of various existing methods by considering their applications domain, an original image segmentation system is proposed. The originality of the developped system lies in the adaptation of processings considering the local context with the minimum of a priori knowledges. The system is composed of three processing modules. The first one allows to analyze the image at two levels. On the one hand, the first level enables to determine the global context of the image to process (image mainly composed of uniformed regions and textured ones) in order to adapt posterior processings and on the other hand, to localize textured and uniformed areas. The second level concerns the local analysis of the image to segment in order to characterize each detected areas considering classical texture attributes that are significant (derived by a statistical analysis) and attributes we have defined. These complementary parameters are determined from a texture model derived from the Wold decomposition of the autocovariance function. They enable to get some informations on the type of texture (stochastic, deterministic) and on its granularity (macroscopic, microscopic). This more precise analysis of a textured region allows to make the choice of the segmentation method easier and secondly to adapt the analysis window size of the region to segment. The second module triggers the segmentation method which is adapted to the local context of the image by using an unsupervised classification method that we have developped. Finally, the third module enables to fusion either the results of several segmentation methods of a same image or results of each band of a multi-components image. The developped fusion method is based on a genetic approach combining the segmentation results by taking into account an evaluation criterion. The system has been validated on different types of images (synthetic and remote sensing).
Document type :
Complete list of metadatas
Contributor : Christophe Rosenberger <>
Submitted on : Thursday, December 6, 2007 - 3:39:08 PM
Last modification on : Wednesday, May 16, 2018 - 11:24:09 AM
Long-term archiving on: : Monday, April 12, 2010 - 6:27:52 AM


  • HAL Id : tel-00194453, version 1



Christophe Rosenberger. Mise en oeuvre d'un système adaptatif de segmentation d'images. Interface homme-machine [cs.HC]. Université Rennes 1, 1999. Français. ⟨tel-00194453⟩



Record views


Files downloads