摘要：The metric spaces model formalizes the similarity searchconcept in nontraditional databases. The goal is to buildan index designed to save distance computations when an-swering similarity queries later.A large class of algorithms to build the index are basedon partitioning the space in zones as compact as possible.Each zone stores a representative point, called center, anda few extra data that allow to discard the entire zone atquery time without measuring the actual distance betweenthe elements of the zone and the query object. The way inwhich the centers are selected affects the performance of thealgorithm.In this paper, we introduce two new center selection tech-niques for compact partition based indexes. These tech-niques were evaluated using the Geometric Near-neighborAccess Tree (GNAT). We experimentally showed that theyachieve good performance
