Indexing Metric Spaces with M-Tree

M-tree is a dynamic access method suitable to index generic “metric spaces”, where the function used to compute the distance between any two objects satisfies the positivity, symmetry, and triangle inequality postulates. The M-tree design fulfills typical requirements of multimedia applications, where objects are indexed using complex features, and similarity queries can require application of time-consuming distance functions. In this paper we describe the basic search and management algorithms of M-tree, introduce several heuristic split policies, and experimentally evaluate them, considering both I/O and CPU costs. Results also show that M-tree performs better than R∗-tree on highdimensional vector spaces.