Discriminating the Machine-Printed and Hand-Written Words Based on Legibility

Abstract

Discrimination of machine-printed and hand-written words is deemed as a major problem in the recognition of the mixed texts. To present a new method to distinguish between machine-printed words and hand-written words using a novel statistical feature on base legibility and discriminator threshold are objectives of this study. Because of the hand trembling, sudden uncontrollable movement of hand and sudden pen shift on the paper, machine-printed words are more legible than hand-written words. The feature is extracted using the Freeman chain code as they are focused on measurement of words legibility. The obtained quantity, which is introduced in this work for the first time, could be a distinguishing criterion for machine-printed words from hand-written. Practically, our method is applied to a mixed and unrefined Farsi database which includes the two above typologies of words. Removing machine-printed words from database and constructing a pure hand-written Farsi words is the other objective. Determining the threshold level, the accuracy rate of the method employed was calculated to be over 96.02%.