Patent US 4,903,312

An input, unknown character is optionally read to form a binary character image, and a plurality of directionality codes are assigned to pixels defining the contour of the character image selectively to define a code-assigned character image, which is then subdivided into a plurality of sub-regions in the form of mesh such that each sub-region includes a substantially equal number of assigned codes. Then, a histogram of the codes is defined for each of the sub-regions to form a collection of histograms for the input character. The input character is compared with each of registered, known characters, each having a collection of similar histograms, and the one having a greatest similarity is selected to identify the input character. The similarity is determined by calculating a distance between the histograms of the input character and the histograms of the each of the registered characters. Preferably, the binary character image is subjected to line-broadening processing prior to the step of assigning the codes.