Automatic estimation of the LVQ-1 parameters. Applications to multispectral image classification

Nearest neighbor rules are widely used nonparametric classifiers in pattern recognition. The main drawbacks of these rules are related to their computational effort. In that sense some techniques have been proposed to select a reduced and representative reference set from the original training set. Adaptative learning techniques can be used successfully to reduce the reference set size. The main drawback of these techniques is the required accurate selection of the parameters involved. In this paper we propose two algorithms to estimate the parameters involved in the LVQ-1 learning and we show that we can get a high accuracy in the 1-NNR classification using the reference set learned by LVQ-1. The proposed algorithms can be easily extended to others adaptative learning methods.