Ontology Based Object Learning and Recognition

Abstract : This thesis deals with the problem of complex object recognition. The proposed approach takes place in the conceptual framework of cognitive vision. This thesis shows how an object categorization system is set up in three phases.
The knowledge acquisition phase consists of acquiring domain knowledge as a taxonomy/partonomy of domain classes. It also consists of acquiring the visual description of these domain classes. This description is driven by a visual concept ontology composed of several types of concepts (spatial concepts and relations, color concepts and texture concepts). Each visual concept of the ontology is associated with low-level features and algorithms. The visual concept ontology stands as a user-friendly interface between expert knowledge and image processing level.
The learning phase results in a set of visual concept detectors. The role of a visual concept detector is to detect visual concepts used during knowledge acquisition in any image. A visual concept detector is obtained by training Support Vectors Machines with features extracted in segmented image samples labeled by visual concepts.
The categorization phase uses both the acquired domain knowledge and the visual concept detectors obtained during the learning phase. Domain knowledge is used to generate hypotheses which have to be verified in the image by visual concept detection in automatically segmented images. The categorization result is composed of the objects recognized in the image with their visual description.
The approach has been applied to the problem of semantic image indexing and retrieval.
Document type :
Theses
Complete list of metadatas

Cited literature [122 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00327542
Contributor : Service Ist Inria Sophia Antipolis-Méditerranée / I3s <>
Submitted on : Wednesday, October 8, 2008 - 4:22:18 PM
Last modification on : Saturday, January 27, 2018 - 1:30:50 AM
Long-term archiving on : Friday, June 4, 2010 - 12:21:59 PM

Identifiers

  • HAL Id : tel-00327542, version 1

Collections

Citation

Nicolas Maillot. Ontology Based Object Learning and Recognition. Interface homme-machine [cs.HC]. Université Nice Sophia Antipolis, 2005. Français. ⟨tel-00327542⟩

Share

Metrics

Record views

430

Files downloads

1189