Efficient similarity search in feature spaces with the Q-Tree

Abstract

In many database applications similarity search is a typical operation that must be efficiently processed. A class of techniques to implement similarity queries which has demonstrated acceptable results includes the use of a nearest neighbor search algorithm on top of a multidimensional access method. We propose a newalgorithm based on the space organization induced by the Q-tree. Specifically, a metric is used to discard internal zones of a cube that have been previously extracted from it. Our experiments based on both synthetic and real data sets showthat our approach outperforms other competitive techniques like R*-tree, M-tree, and Hybrid-tree. © Springer-Verlag Berlin Heidelberg 2002.