The use of threshold logic in character recognition

The fact that many complex counting and decision functions can be realized quite simply with threshold gates suggests that they may be used to considerable advantage in problems of character recognition. A simplified recognition problem is considered involving the identification of any one of 12 letters when it is superimposed on an m × n matrix. Translation, stretching, and compression of the letter are permitted. It is shown that the number of threshold gates required increases linearly as do the dimensions of the matrix with about 300 gates being necessary for a 20 × 20. On such a matrix, several hundred thousand configurations of the 12 letters can be correctly identified with each pattern being insensitive to varying degrees of "noise." A threshold gate having the necessary fan power for this application is described together with its implementation in a small experimental model. Extensions of the methods to include rotation and magnification are discussed.