A novel spatial index for case based geographic retrieval

The case based geographic retrieval is a kind of content based multimedia retrieval for the vector-like data. The goal of the retrieval system is to find the geographic scenes from the geographic information system, which are similar to the hand-drawn query scene according to the topological and shape characteristics. In this paper, a spatial index structure and a two-stage retrieval algorithm are proposed to support the case based geographic retrieval, which cannot be performed by traditional spatial index trees. The spatial index structure in this paper is composed of two index trees. One is the extended R-Tree, which indexes the spatial scenes based on the spatial distance relationship. The other is the spatial relationship index tree, which clusters the spatial relationship feature vectors of the scenes in an R*-tree. Bidirected pointers from the leaf nodes of the spatial relationship index tree to the scene nodes in the ER-Tree connect the two index trees. The two-stage retrieval algorithm searches the nearest feature points to the query's feature point in the spatial relationship index tree firstly. Then the multi-scale result scenes can be found in the ER-Tree. Experiments show that the spatial index structure has reasonable insertion speed and efficient retrieval speed. The new spatial index structure also provides multi-scale and multi-granularity geographic retrieval.