Automatic recognition of low-quality printed characters using analogue techniques

After comment on the obvious disparity between the print quality accepted as normal for human reading and the much higher quality needed for machine reading, a system is proposed and results of a simulation study presented for reducing this gap. The system employs a modified form of auto-correlation to extract characteristic properties from which specimen unknown characters can be identified by a simple recognition logic. In order to avoid the problem of representing the continuous grey-tone density scale of low quality print as a two state (binary) variable, analogue techniques are proposed for use in the system. Using a computer simulation of the proposed system a limited number of printed samples of exaggerated quality distribution are used to explore the limits of performance of the system. Numerical results are given and factors effecting performance discussed.