DSACL+-tree: A dynamic data structure for similarity search in secondary memory

Abstract

Metric space searching is an emerging technique to address the problem of efficient similarity searching in many applications, including multimedia databases and other repositories handling complex objects. Although promising, the metric space approach is still immature in several aspects that are well established in traditional databases. In particular, most indexing schemes are not dynamic. From the few dynamic indexes, even fewer work well in secondary memory. That is, most of them need the index in main memory in order to operate efficiently. In this paper we introduce two different secondary-memory versions of the Dynamic Spatial Approximation Tree with Clusters (DSACL-tree from Barroso et al.) which has shown to be competitive in main memory. These two indexes handle well the secondary memory scenario and are competitive with the state of the art. But in particular the innovations proposed by the version DSACL+-tree lead to significant performance improvements.The resulting data structures can be useful in a wide range of database application. © 2012 Springer-Verlag.