Computer System for Support of Humans With Damaged Sight: Subsystem for Optical Character Recognition of Printed Cyrillic Text

Dejan Gorgevik, Dragan Mihajlov, Ljubomir Josifovski

Abstract: A subsystem for optical character recognition of printed Macedonian Cyrillic text as a part of a system for supporting humans with damaged sight is presented. The system includes printed Cyrillic text recognition, and its printing on a Braille printer or speech synthesis. Various commercial programs for optical character recognition had been studied, but they showed to be inappropriate for implementation in the system. Therefore a program for optical character recognition was developed. Various types of classifiers like: simple overlapping of binary matrixes, neural network based perceptron, back-propagation and adaptive logical networks have been tested. Comparative analysis considering their accuracy, speed, resource requirements, and sensitivity to noise and deformations have been carried out.

Keywords: Cyrillic text, optical character recognition, adaptive logical network, neural network