In this paper, an optical character recognition system for hand-writte
n rotated digits in land registry maps is presented. It is based on a
neural network and trained by a constructive learning rule, the Hyperb
ox Perceptron Cascade (HPC). The HPC classifier can design complex, po
ssibly nonconvex, disjoint, and bounded decision regions and treat the
rejection problems of outliers and unanticipated patterns, which woul
d otherwise tend to be classified positively in an incorrect class. We
use ''shape features'' and a novel approach to select the most promis
ing features to attain a low generalization error. The numerous experi
ments show that a subset of 24 of the 46 features obtains a good class
ifier with a high rate of correct classification and a low rate of rej
ection.