An Analog Method for Character Recognition

Abstract

A method for character recognition which is capable of an analog implementation has been studied by simulation on asymbols digital computer. In essence, this method involves maximizing the cross-correlation value between the unknown character and a set of average characters, there being one average character for each allowed character class. An average character is represented by a two-dimensional function. The value of this function at a point is the probability of occurrence of a mark at that point for the character class represented by the average character. Negative weights are given to areas of low probability in each average character to improve discriminability. The simulation results indicate that this method is applicable to the recognition of machine printing, and perhaps to the recognition of constrained hand printing. The method can be implemented in an economical manner using electro-optical techniques.