Recognition of handwritten numerical characters for automatic letter sorting

A prototype of a postal code number reader capable of recognizing handwritten arabic numerals has been developed and tested successfully. Recognition of handwritten arabic numerals requires sufficient system flexibility to cope with both unlimited variations in character shape and a large number of writing instruments. This flexibility has been achieved by combining the use of a stored recognition table and a hardware microprogram. The recognition logic obtained by computer simulation is immediately stored in the core memory as a table. The numerals are recognized by extracting a sequence of the geometrical features in horizontal zones of the character after normalization of the height of the character and the width of the strokes. The use of sequential recognition logic helps reduce the capacity of core memory required to obtain the desired high rate of recognition. The correct recognition rate of a single digit for a large sample of letters averaged 95 percent. For a three-digit sorting with 13 stackers the sorter incorrectly recognized approximately 0.1 percent, a rate of error consistent with manual sorting.