Spatial Indexing by Virtual Bounding Rectangles for High-Dimensional Data

Abstract

This chapter presents a new indexing scheme, the Subspace Coding Method (SCM), that offers high performance similarity retrieval. Indices in the R-tree family generate tree structure using MBRs (Minimum bounding rectangles) or MBSs (Minimum bounding spheres). Our proposed method is based on tree structure using MBRs (i.e. the R-tree, the R*-tree, the X-tree and so on), and newly introduced the notion of VBR (Virtual Bounding Rectangle). VBRs are rectangles which contain and approximate MBRs. Importantly, the notion of VBR is orthogonal to any other method in the field of spatial search and is introduced into any spatial indices using MBRs. Furthermore, we have introduced the Subspace Coding Method, which can compactly represent VBR. The performance evaluation shows the superiority of our method for high-dimensional data.