Strategies for efficient incremental nearest neighbor search

Abstract

Algorithms for determining the m nearest neighbors of a query point in k-dimensional space can be inappropriate when m cannot be determined in advance. We recast the m nearest neighbor problem into a problem of searching for the next nearest neighbor. Repeated invocations of an incremental searching algorithm allow an arbitrary number of nearest neighbors to be determined. It is shown that incremental search can be implemented as a sequence of invocations of a previously published non-incremental algorithm. A new incremental search algorithm is then presented which finds the next nearest neighbor more efficiently by eliminating redundant computations. Finally, the results of experimental computer runs comparing the two approaches are presented.