On the Reading of Postal Codes by Machine

Abstract

The paper describes the current position of a long-term research project for the sorting of mail by automatic reading of the postal codes. Of the four sub-problems, i.e. manipulation (of the envelopes), extraction (of the information), location (of the code), and recognition (of the code once located and extracted), the fourth alone is considered; in particular, attention is concentrated on machine-printed characters. It is necessary to describe the characters in terms of ‘features’, which will conceal differences in style but exaggerate differences in kind. This leads to the concept of a ‘three-layer’ machine, possessing quite remarkably brain-like qualities. The ideal feature list, conceived as a basic set of vectors in an abstract hyper space, has very interesting properties.