An evaluation of the WPE algorithm using tangent distance

Weighting prototype editing (WPE) is a novel approach to editing a given set of prototypes so that the resulting set can outperform the original one in terms of the nearest neighbor (NN) classification accuracy. This technique is applied in this work along with an interesting dissimilarity measure between pixel maps, known as tangent distance (TD). Experiments on the USPS handwriting digits benchmark corpus are presented, with results showing the capability of the WPE to improve the already good results based on TD NN classification.