NNH: Improving Performance of Nearest-Neighbor Searches Using Histograms Liang Jin, Koudas Nick, and Chen Li   View Paper (PDF)   Return to Session 7: Query Processing Techniques for Spatial Databases Abstract Efficient search for nearest neighbors (NN) is a fundamental problem arising in a large variety of applications of vast practical interest. In this paper we propose a novel technique, called NNH ("Nearest Neighbor Histograms"), which uses specific histogram structures to improve the performance of NN search algorithms. A primary feature of our proposal is that such histogram structures can co-exist in conjunction with a plethora of NN search algorithms without the need to substantially modify them. The main idea behind our proposal is to choose a small number of pivot objects in the space, and pre-calculate the distances to their nearest neighbors. We provide a complete specification of such histogram structures and show how to make use of the information they provide towards more effective searching. In particular, we show how to construct them, how to decide the number of pivots, how to choose pivot objects, how to incrementally maintain them under dynamic updates, and how to utilize them in conjunction with a variety of NN search algorithms to improve the performance of NN searches. Our intensive experiments show that nearest neighbor histograms can be effi- ciently constructed and maintained, and when used in conjunction with a variety of algorithms for NN search, they can improve the performance dramatically. ©2005 Association for Computing Machinery