An exact fast nearest neighbor identification technique

A novel fast algorithm for the determination, among a M-dimensional set of points, of the neighbor nearest to a given point by using the Euclidean metric space is presented. First, a region which contains the unknown point and at least one neighbor is found by using an ordered search procedure for the unknown point from O/sub i/, i=1,. . .,g reference points. Second, the region found is modified such that the nearest neighbor to the given point will be inside it. The nearest neighbor is then determined at the expense of a small number of distance calculations. The number g and the positions of the reference points are determined properly taking into account the structure of the initial set of points and always keeping the search region to a minimum. The preprocessing load is considered to be moderate. The speed and accuracy of the algorithm are demonstrated experimentally.<>