Thèse soutenue

Vers une plate-forme de vision cognitive pour l'interprétation sémantique d'images : application à la reconnaissance d'organismes biologiques
FR  |  
EN
Accès à la thèse
Auteur / Autrice : Céline Hudelot
Direction : Monique Thonnat
Type : Thèse de doctorat
Discipline(s) : Informatique
Date : Soutenance en 2005
Etablissement(s) : Nice
Ecole(s) doctorale(s) : École doctorale Sciences et technologies de l'information et de la communication (Sophia Antipolis, Alpes-Maritimes)

Résumé

FR  |  
EN

This work deals with cognitive vision and in particular semantic image interpretation. One of the main challenges of cognitive vision is to develop a flexible, adaptable system, capable of performing complex image analysis tasks and of extracting information from various scenes and images. The objective of this thesis is to cope with this challenging problem by the design of a reusable and generic cognitive vision platform for the complex problem of semantic image interpretation. We are interested in both the cognitive issues and the software engineering ones involved in the design of such a platform. The proposed cognitive vision platform is a unified environment which proposes generic and reusable tools for the design of semantic image interpretation systems. The semantic image interpretation problem is complex and can be divided into three more tractable sub-problems: (1) the semantic interpretation, (2) the problem of mapping between the high level representations of physical objects and the sensor data extracted from images, (3) the image processing problem. To manage and separate the different sources of knowledge and reasoning, we propose a distributed architecture based on the cooperation of three Knowledge Based Systems (KBS). Each KBS is highly specialized for the corresponding sub-problem of semantic image interpretation. For each sub-problem, we define a dedicated engine and a unified knowledge representation model. The implementation of the cognitive vision platform has been made with the LAMA platform, a software platform for the development of knowledge based systems designed in the ORION team. To validate our cognitive vision platform, we have chosen a real world application: the early diagnosis of plant diseases. In particular, we have focused on the rose leaf diseases in greenhouses. This work has been made in cooperation with INRA (French National Institute for Research in Agronomy).