A system for the automatic recognition of patterns

The paper describes a new method for the automatic recognition of patterns. The method may be applied to any form of spatial pattern, but in the present instance, patterns consisting of line figures are considered. The pattern is presented to a flying-spot scanner connected to a digital computer. The shape of the pattern is analysed and a statement is prepared describing the basic features of the pattern. The pattern is then recognized by comparing this statement with a number of others already stored in the computer which relate to named patterns. Patterns are recognized independently of the angle at which they are presented to the scanner, and may be of any size provided that limits imposed by the resolution of the scanner are not exceeded. The average time to recognize a character is 60 seconds with the system programmed on a medium-speed computer. Special-purpose equipment built to perform certain of the stages of the process, together with the use of higher-speed computers now envisaged, will reduce this time by at least a thousandfold. If a new pattern is presented to the machine it will indicate its inability to recognize the pattern, but by giving the machine the name of the pattern, it may become one of the standard patterns which it can subsequently recognize. All the patterns recognized by the machine are hand-drawn and consist of such symbols as the capital letters of the alphabet and numerals, although the system is in no way limited to any special set of characters. Using exactly the same method but with an increase in the degree of complexity, it will be possible for machines to read handwriting. Special allowances are made for imperfections in the patterns, including breaks and general ill-definition. Where there is some confusion and an unknown pattern resembles two or more of the standard patterns, the relative degrees of similarity of the unknown to each of these standard patterns is printed out by the machine.

[1]  C. K. Chow,et al.  An optimum character recognition system using decision functions , 1957, IRE Trans. Electron. Comput..