TEXT RECOGNITION USING ADAPTIVE SOFTWARE.

Abstract

A unique system for entering optically scanned printed text into a computer as coded information is presented. Given an arbitrary document to encode, the system first extracts one or more examples of each symbol on the page. These 'prototypes' are postanalyzed to detect and separate merged characters, then presented to an operator who keys in identity codes. The remaining text characters are assigned the identifies of the prototypes that they resemble. Dictionary searching is also included as an alternative method of encoding prototypes and reducing the need for manual keying. The method provides low-cost input of text from books and journals, whose fonts are not accepted by conventional QCR machines, as well as from typed office documents.