Improving the R*-tree with outlier handling techniques

10 years 10 months ago
Improving the R*-tree with outlier handling techniques
The R*-tree, as a state-of-the-art spatial index, has already found its way into commercial systems like Oracle. In this paper, we aim at improving query performance of the R*tree. We focus on five widely used spatial queries: range query, aggregation query, nearest neighbor query, skyline query, and join query. The idea is to store outlier objects in internal tree nodes. The new structure is named the RO tree. Here an outlier is an object which is located far from other objects or has large extent (we consider both point objects and objects with extent). If such objects are stored at higher levels of the tree, the lower-level nodes have smaller minimum bounding rectangles and thus the index performs better. To support the dynamic nature of the index, several structural and algorithmic changes are needed. The paper discusses these changes. In particular, we show how to identify and handle the outlier objects during page overflow/underflow, using gain/loss metrics. Extensive experiment...
Tian Xia, Donghui Zhang
Added 11 Nov 2009
Updated 11 Nov 2009
Type Conference
Year 2005
Where GIS
Authors Tian Xia, Donghui Zhang
Comments (0)