A method for the optimal design of a class of pattern recognition systems

摘要

Easily implemented measurements which allow an equal variance in all portions of a pattern are made on single rows and columns of a two-dimensional binary image of the pattern. An optimal minimum set of measurements for each size binary image is determined by solving a prime implicant chart. Application of an economic principle determines the best system. A simple example is presented to demonstrate the method.The best system for a set of 52 characters is determined using the method. Results indicate that a maximum allowable variation in any letter is 30.9 per cent, regardless of binary image size. An hypothesis relating character set size and character similarity is stated.