Parallel recognition of idealised line characters

Summary

The aim is to design a machine which is able to learn a number of idealised characters and to recognise them, irrespective of their size, position and context on an infinite retina. If the number of characters which such a machine can possibly learn to recognise is astronomical, it is not practical to use separate templates for every possible character. It is more economical to use, instead, templates for various parts, called features, of characters. In recognising a number of characters simultaneously, without scanning, the question arises of how to tell which feature belongs to which character of figure on the retina. In particular, if a given character is not present but all its features are included in nonsense figures simultaneously present on the retina then the machine must not indicate the presence of the given character. The technique which overcomes this difficulty employs overlapping features which must be mutually consistent for recognition. This consistency technique is assessed by comparison with a more conventional technique, and the work is restricted to closed line characters which are not subject to deformations or mutilations.