Approximate nearest neighbour search with the Fukunaga and Narendra algorithm and its application to chromosome classification

Abstract

The nearest neighbour (NN) rule is widely used in pattern recognition tasks due to its simplicity and its good behaviour. Many fast NN search algorithms have been developed during last years. However, in some classification tasks an exact NN search is too slow, and a way to quicken the search is required. To face these tasks it is possible to use approximate NN search, which usually increases error rates but highly reduces search time. In this work we propose using approximate NN search with an algorithm suitable for general metric spaces, the Fukunaga and Narendra algorithm, and its application to chromosome recognition. Also, to compensate the increasing in error rates that approximate search produces, we propose to use a recently proposed framework to classify using k neighbours that are not always the k nearest neighbours. This framework improves NN classification rates without extra time cost. © Springer-Verlag Berlin Heidelberg 2003.