Evaluation of strategies for multiple sphere queries with local image descriptors - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2006

Evaluation of strategies for multiple sphere queries with local image descriptors

Résumé

This paper was devoted to the study of similarity search in a collection of points in high-dimensional space. We focused on sphere queries and on SR-trees as a structure for accelerating sphere queries. Our first contribution was to experimentally show that when the dimension of the space is moderate, i.e. between 8 and 30, the tree structure indeed performs well: it exhibits a significant gain wrt to sequential scan even in a 29-dimensional space. Our second and main focus was on the improvement of multiple sphere queries. The latter are useful in a large number of data mining applications as well as for image retrieval by content. As shown by Braunmuller and al. for X-trees and k-NN queries, we showed that for sphere queries on an SR-tree, I/O cost is significantly reduced by processing simultaneously all queries and accessing only once each relevant node tree for all queries. Last, and this is our main contribution, an extensive performance evaluation showed that when the points in the multiple query are clustered, then CPU time can be saved as well, in processing all queries simultaneously, by avoiding distance computations. For this we enlarged the set of lemmas based on the triangle inequality proposed by Braunmller and al. Our performance study relied on a real data collection drawn from a content based image search application as well as on four synthetic data sets. These trends were confirmed by querying a database of images with queries on objects of interest.
Fichier non déposé

Dates et versions

hal-01125102 , version 1 (06-03-2015)

Identifiants

  • HAL Id : hal-01125102 , version 1

Citer

Nouha Bouteldja, Valérie Gouet-Brunet, Michel Scholl. Evaluation of strategies for multiple sphere queries with local image descriptors. IST/SPIE Conference on Multimedia Content Analysis, Management, and Retrieval, San Jose CA, USA, Jan 2006, X, France. pp.A1-12. ⟨hal-01125102⟩

Collections

CNAM CEDRIC-CNAM
20 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More