| A Multi Line Optical Character Recognition system (MLOCR), as defined by the U.S. Postal Service, is a machine that reads the addresses on envelopes or parcels and automatically sorts them into the correct bins according to their destinations. This dissertation describes the hardware design of the Optical Character Recognition (OCR) subsystem used in a small, cheap, very fast MLOCR.; The OCR subsystem receives as input a single character which has already been scaled to standard size, rotated to a standard orientation, and binarized. The OCR output is a list of one or more possible identifications of the character, or a '?' indicating that it cannot reliably identify it. The developed OCR subsystem can recognize a single character in less than 200 microseconds average, has an error rate of less than 1% (correct single character, correct character in list, or "?" returned {dollar}>{dollar} 99%), '?' is returned less than 5% of the time, and an ambiguous list returned less than 30% of the time.; The OCR subsystem consists of a template tree subsystem and a template matching subsystem. This dissertation describes the implementation of the template matching subsystem. Five custom mixed signal integrated circuits and a printed circuit board (PCB) were developed. The subsystem is capable of giving a distance calculation between an input character and a template in less than 2 microseconds, which allows a character to be recognized in less than 200 microseconds by the template tree subsystem. |
