Real time optical character recognition is required in a large number of machine vision applications. This proposal describes a unique method of reading characters, even those presented in a misoriented manner. It takes the "connectivity" concept found in artifical intelligence software and converts it to hardware. Starting with this hardware connectivity board, additions are made to it to permit the in-line reading of characters. These additions and changes will include transmission of the image to a vision module for re-orientation; transient storage of the image (e.g., a full page) viewing each character from the four cardinal directions, and access to blob connection points. Each character is looked upon as a connected object which is constructed from a series of blobs. This upgraded connectivity board can itself go a long way towards the unique determination of each viewed character by extracting two new features, the number of lakes in the character and the directions of its bays and by comparing them to those catalogued in a lookup table. There are a few sets of characters that have identical bay and hole features (e.g., C,F,G). In such cases, the machine vision module used in association with this connection concept, measures each character's features (centrold, area, perimeter) again in real time, to determine the character uniquely. Font independence is possible by a later addition of a serif removal circuit. The technique is applicable to both capital and small letters and can be extended to Chinese, Islamic, and more complex pictograms.The potential commercial application as described by the awardee: The research may lead to the development of a technique for fast facsimile transmission, a CAD/CAM drawing reader, a license plate reader, production line serial number reading, document and page readers, wired community, home computers, and Chinese, lslamic, or Vietnamese fonts.
