A method for recognition of hand printed characters.

Abstract

A practical and effective preprocessing and recognition method for the pattern recognition machine is presented. The pattern recognition machine considered is simulated on a digital computer. Samples of five pattern classes which consist of hand-printed characters, A, B, C, D, E, are used in the experiments. The results are found very satisfactory. A preprocessing method, called Sequential and Centroid Transformation (SCT), is developed to extract the characteristic features from patterns. A recognition method, called Least Squares Recognition Algorithm, based on the properties of the SCT method is developed and is shown to satisfy the principle of least squares. One hundred samples which are collected randomly from several persons are used as input testing patterns. The samples used are not only translated, rotated, and dilated or shrunk but also found to contain a reasonable amount of random transformations such as change of proportions, bending of lines, etc. Nevertheless, the recognition rate is as high as 95 per cent.