Abstract
In this paper the divide-and-conquer approach to the two-dimensional closest-pair problem is studied. A new algorithm is proposed by which a closest pair, i.e. a pair of points whose distance δ is smallest among a set of N points, is found in θ(N) expected time when the points are drawn independently from the uniform distribution in a unit square. The worst-case running time of the algorithm is θ(N log2 N). The method is to project the points onto one of the coordinate axes, and to compute an initial guess for the smallest distance δ by considering the [N/2] pairs of successive projected points. The shortest of these pairwise distances is a good approximation for the final δ. It is then used in the subsequent merge phases of the divide-and-conquer algorithm to keep the average work minimal. A modification of the basic algorithm guarantees θ(N) performance in the average case and θ(N log N) performance in the worst case.
C.R. Categories: