Rejection strategies and confidence measures for a k-NN classifier in an OCR task
Joaquim Arlandis, Juan-Carlos Pérez-Cortés, Jesús Cano · 2003
In handwritten character recognition, the rejection of extraneous patterns, like image noise, strokes or corrections, can improve significantly the practical usefulness of a system. In this paper a combination of two confidence measures defined for a k-nearest neighbors (NN) classifier is proposed. Experiments are presented comparing the performance of the same system with and without the new rejection rules.