Automated Reading of the Printed Page

Abstract

This paper is primarily concerned with the automated reading of the printed page resulting in a sequence of character codes which can be further processed to make the information available to a blind person in the form of Grade II Braille, spelled speech, or synthesized speech. Heuristics are described for automatic threshold determination, font-size determination, line and character acquisition, contour tracing, and the recognition of punctuation and characters. As the output of the reading machine for the blind is to be absorbed directly by a human, the specifications for a page reader were that the speed should be approximately equal to normal speaking rates and that the reader should make somewhat fewer mistakes than a human reader would make. The actual speed achieved was approximately 75 words per minute, which does not quite meet the speed requirement. However, the error rate specification has been met with a measured error rate of .07 per cent.