Abstract

Elementary Perceiver And Memorizer (EPAM) is one of the few existing general models of human perception. It has been successful in simulating and describing many human cognition phenomena. In this thesis, the role of EPAM is studied in terms of an OCR machine. The model is described in some detail. Its strengths and shortcomings have been pointed out. Some suggestions are provided for its shortcomings. Two of the suggestions are implemented and tested in this thesis. In EPAM, objects of any complexity level could be described. In this experiment the primitive objects of the EPAM model are defined as characters, and words of the English language are defined as the upper bound on the complexity of the objects. In this way the model is tested under: (1) One level of object complexity. (2) A large base of objects with a variable length feature set. As such, to test the modification's effect on the behaviour of the model, EPAM is used as a postprocess to a character segmentation and recognition subsystem. The experiment is tested with 30 fonts of different typefaces with sizes ranging between 7 points to 12 points. For the segmentation subsystem a line, word, and character segmenter is provided. A Feature extraction methodology and a recognition model are also provided for the character recognition subsystem. EPAM is trained with 23692 words and tested with the results of the character testing font set.