A Character Recognition Application of an Iterative Procedure for Feature Selection

A simple yet powerful technique for selecting features to be used in a pattern recognition system has been devised and applied to an eight-class character recognition problem using a set of 19 000 typed character samples as a data base. High-order joint probabilities have been directly estimated from the data base, thus making it possible to take into account in the feature selection process the existence of statistical dependencies among features to a greater extent than has been done in previously reported work.