Skip to Main content Skip to Navigation
Theses

Contribution à une méthodologie d'intégration de connaissances pour le traitement d'images. Application à la détection de contours par règles linguistiques floues

Abstract : In this thesis, we propose a contribution to a methodology for knowledge integration in image processing that we apply to edge detection. We outlined this methodology under the base of the theory of fuzzy information granulation, recently suggested by L A. Zadeh. Thus, we built a contextual model, for an edge detection application, that incorporates knowledge about the elements relating to the acquisition system and that takes into account the imprecision of real contours. This model is inserted in a fuzzy image processing paradigm. This document is articulated in four parts. First, we present our point of view about an industrial vision system. Next, we introduce linguistic modeling as the theoretical base of approximate reasoning as same as the theory of the fuzzy information granulation. Second, we start with a generalization of image processing applications using fuzzy set theory through the paradigm of fuzzy image processing. Then we particularize this paradigm for fuzzy rules system case in order to construct our edge detection operator FRED, (Fuzzy Reasoning Edge Detection). FRED consists of five modules : the extraction of properties, the fuzzification, the reasoning, the defuzzification and the contextual model related to the knowledge base. On the ground of fuzzy information granulation, we presented a methodology to construct the contextual model, implying the stages of granulation, attribution and valuation. Third, we detail the implementation of two operators of detection of contours on the basis of FRED. Although these two operators use the same generic structure and have the same objective, edge detection, they translate nevertheless two different contexts. The results of these operators are then aggregated with a specific fusion operator that we have especially developed. And fourth, we validate the implementations of our two operators\ using Fram and Deutsch criteria on computergenerated images. These criteria reflect the robustness of the operator with respect to the noise and the continuity of the contour's segment. An application concerning a renal arteriography edge detection process ends this document.
Document type :
Theses
File URL :
http://docnum.univ-lorraine.fr/prive/SCD_T_2000_0091_PEREZ_ORAMAS.pdf
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-01746466
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 10:38:22 AM
Last modification on : Thursday, December 17, 2020 - 2:06:10 PM

Identifiers

  • HAL Id : tel-01746466, version 1

Collections

Citation

Oliver Perez Oramas. Contribution à une méthodologie d'intégration de connaissances pour le traitement d'images. Application à la détection de contours par règles linguistiques floues. Autre. Université Henri Poincaré - Nancy 1, 2000. Français. ⟨NNT : 2000NAN10091⟩. ⟨tel-01746466⟩

Share

Metrics

Record views

12