Abstract

Large image and spatial databases are becoming more important in applications such as image archives and spatial data mining. In similarity searches, e.g., in image and multimedia systems providing content based search, high dimensional spaces are common. Thus, spatial indexing methods which can effectively deal with large data sets of high dimensionality are of great interest. In this paper we present a new index structure which is based on regular spatial decomposition but which uses a unique set of constraints to eliminate useless nodes that can result from a skewed spatial distribution of objects. The index structure can be formally defined and is unique for a given data set. It is efficiently update-able and bounds on the number of nodes and the height of the tree can be proved. Also bounds on the expected height of the tree can be given if certain mild constraints on the spatial distribution of objects can be given. Empirical evidence shows that the PK-tree outperforms the existi...

Citations