Description
Date depot: 1 janvier 1900
Titre: Analyse et la recherche d'évènements à partir de séquences d'images dans un contexte distribué
Directrice de thèse:
Nozha BOUJEMAA (Inria-Paris (ED-130))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
The predominant model of today's major search engines is based on centralized or clustered architectures. We saw however in the last decade an increasing interest for distributed information retrieval systems, usually in P2P structured networks. Regardless ongoing debates about bandwith consumption, such systems theoretically offer the possibility to cope with web-scale document collections by distributing the indexing and querying load over large networks of collaborating peers. They also represent a serious alternative regarding data confidentiality and information monopoly issues. Almost all proposed distributed information retrieval architecture are purely text-based. First content-based distributed retrieval methods only appeared three years ago notably with the Metric-CHORD structure. Such methods are very promising but are for now highly limited, with very simple search paradigms like range queries, using classical indexing structures like M-trees. There is thus a wide research investigations area to develop distributed solutions for other content-based search paradigms such as distributed similarity search, classification or navigation.
The work could be focused at the beginning on the problem of a distributed KNN classifier over image collections.
Doctorant.e: Trad Mohamed Riadh