High Dimensional Feature Indexing Using Hybrid Trees

Feature based similarity search is emerging as an important search paradigm in database systems. The technique used is to map the data items as points into a high dimensional feature space which is indexed using a multidimensional data structure. Similarity search then corresponds to a range search over the data structure. Traditional multidimensional data structures (e.g., R-tree, KDB-tree, grid les) are of limited use for feature indexing since (1), their performance deteriorates rapidly with the increase in the dimensionality of the feature space(referred to as the \dimensionality curse") and (2), they do not support range queries based on arbitrary distance functions, a situation that occurs commonly in multimedia feature spaces. This paper introduces the hybrid tree { a multidimensional data structure for indexing high dimensional feature spaces. The hybrid tree combines positive aspects of bounding region (BR)-based data structures (e.g., Rtree, SS-tree, SR-tree) and space partitioning (SP) data structures (e.g., KDB-tree, hB-tree) into a single data structure to achieve search performance more scalable to high dimensionalities than either of the above techniques. Furthermore, the hybrid tree supports range queries based on arbitrary distance functions. Our experiments on \real" high dimensional large size feature databases demonstrate that the hybrid tree scales well to high dimensionality and large database sizes. It signi cantly outperforms both purely BR-based and SP-based index mechanisms as well as linear scan at all dimensionalities for large sized databases.