Fast k-NN classification with an optimal k-distance transformation algorithm

The k-NN classification rule uses information from the k nearest prototypes in order to classify a pattern. In this paper, we improve Warfield's lookup table approach, where the classification problem is reformulated in terms of distance transformations. We propose a new k-distance transformation algorithm using ordered propagation. We show that — using this algorithm — the k-NN classification of F possible patterns in a D-dimensional space has a O(k.D.F) complexity.