UBC Theses and Dissertations
Machine recognition of typewritten characters based on shape descriptors Kanciar, Eugene J.A.
An optical character recognition technique for typewritten letters was developed with application to a personal reading machine for the blind. The feature extraction process defined a character in terms of lines and shapes which are closely related to a person's description of form. The system was developed to identify all upper and lower-case typewritten characters in the alphabet. A letter was described by any combination of seven basic features, usually in a 3 x 3 feature matrix. The extraction of topological (or structural) properties had several advantages; a very small feature dictionary with about 100 code-word entries; quick and simple training procedure for a new font; and, a strong capability to handle character deformities. A separate technique, based on edge examination, was developed to identify characters with prominent diagonal features. Sequential classification was employed throughout the entire system so that recognition was made once a sufficiently unique measure was satisfied. Tests on both repeated characters and typewritten passages produced approximately 97% accuracy when the system was applied to three fonts which varied from a stylized to a serifless print. For a scanning rate of 60 wpm, a recognition speed of two characters per second was achieved. The system was developed on a PDP-12 computer and is fully compatible for realization on a PDP-8 computer with 8K of memory.
Item Citations and Data