18 - Statistical Pattern Recognition

Abstract

This chapter discusses statistical pattern recognition. The principle of the nearest neighbor (NN) algorithm is that of comparing input image patterns against a number of paradigms and then classifying the input pattern according to the class of the paradigm that gives the closest match. The general technique plots the characters in the training set in a multidimensional feature space and to tag the plots with the classification index. Then, test patterns are placed in turn in the feature space and classified according to the class of the nearest training set pattern. It is clear that to achieve a suitably low error rate, large numbers of training set patterns are normally required. This then leads to significant storage and computation problems. Notable among these is that of pruning the training set by eliminating patterns that are not near the boundaries of class regions in feature space, as such patterns do not materially help in reducing the misclassification rate.