Ocr for reading a constraint free hand-written character or the like

Abstract

A character pattern to be recognized is divided into a predetermined number of parallel segments of equal width extending in one direction and character information on respective ones of the segments is read out via a line of photoelectric cells. Two pieces of the character information on two adjacent segments are evaluated together in developing symbolic information indicative of the correlation between two adjacent segment patterns. By the action of simplified feature extraction systems, a plurality of two-dimensional features are established in accordance with all combinations of the symbolic information which have been derived throughout the scanning procedure. The symbolic information is developed particularly with regard to the number of loops within the character pattern, the shape of the left edge portion, the shape of the right edge portion, the shape of the upper edge portion and the shape of the lower edge portion of the character pattern. These combinations of the symbolic information are compared with reference information prestored in a memory of an OCR (optical character reader).