Hyperplane Queries in a Feature-Space M-tree for Speeding up Active Learning
Abstract
In content-based retrieval, relevance feedback (RF) is a noticeable method for reducing the ``semantic gap'' between the low-level features describing the content and the usually higher-level meaning of user's target. While recent RF methods based on active learning are able to identify complex target classes after relatively few iterations, they can be quite slow on very large databases. To address this scalability issue for active RF, we put forward a method that consists in the construction of an M-tree in the feature space associated to a kernel function and in performing approximate kNN hyperplane queries with this feature space M-tree. The experiments performed on two image databases show that a significant speedup can be achieved, at the expense of a limited increase in the number of feedback rounds.
Reference
@inproceedings{jpt-bda07,
author = {Crucianu, M. and Estevez, D. and Oria, V. and Tarel, J.-P.},
title = {Hyperplane Queries in a Feature-Space M-tree for Speeding up Active Learning},
booktitle = {Proceedings of Journ\'ees Bases de Donn\'ees Avanc\'ees (BDA'07)},
date = {October 23-26},
address = {Marseille, France},
year = {2007},
note = {http://perso.lcpc.fr/tarel.jean-philippe/publis/bda07.html}
}
Pdf file (248 Kb)
online BDA website