A Generalized Metric Distance between Hierarchically Partitioned Images

Maude Manouvrier (1) , Marta Rukoz (2) , Geneviève Jomier (1)

1 LAMSADE - Laboratoire d'analyse et modélisation de systèmes pour l'aide à la décision
2 CCPD - Centre d'Informatique Parallèle et Distribuée

This article presents a generalized metric distance, called $\Delta$-distance, between images represented by a tree structure resulting from a recursive image partition. This distance is used to perform content-based image retrieval queries in databases. Delta-distance allows to retrieve images globally similar to a query image. This distance takes into account the location of the image visual features. It can be performed using a multi-level filtering algorithm. Moreover, Delta-distance allows region-based queries. In this case, the resulting images contain quadrants similar to the quadrants selected by the user in the query image or contain quadrants similar to the entire query image. Because it is a generalized distance function, some particular cases of the $\Delta$-distance appear in existing content-based image retrieval systems.

Mots clés: Image database; quad/quin/nona-tree; recursive image decomposition; content-based image retrieval; multi-level filtering; image region similarity

Domaines: Base de données [cs.DB]; Multimédia [cs.MM]

Maude Manouvrier, Marta Rukoz, Geneviève Jomier. A Generalized Metric Distance between Hierarchically Partitioned Images. Proc. of the 6th Intl. Workshop on Multimedia Data Mining - "Mining Integrated Media and Complex Data" (MDM/KDD2005), Aug. 2005. In conjunction with the Eleventh ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining, 2005, Chicago, United States. pp.33-41. ⟨hal-00008820⟩