Character-recognition machines, based on high-resolution flying-spot scanners and sophisticated logic circuitry, promise variable-type-face, high-speed data input directly from machine-printed documents.
Techniques used in the Apple cathode-ray tube are being applied by Philco Corp. to an experimental machine recently demonstrated at the company's Blue Bell, Pa., research laboratories. The Philco recognizer readily reads letters from different kinds of typewriters and even reads simulated typewriter characters drawn by a draftsman.
The techniques de-veloped by Philco are being used in two machines being built for the Post Office Dept. A machine capable of recognizing 25 city-state locations will be delivered to the post office in Washington, D.C., in the spring. An advanced machine, capable of recognizing all 50 states and 250 major cities in the United States, will be built by Philco by sometime in 1963. This machine should be capable of reading 36,000 letters per hour.
Sylvania's Advanced Development Laboratory, Waltham, Mass., is developing a high-speed recognizer based on the flying-spot scanner technique under a company-sponsored program. Although similar to the Philco approach, there are some important differences in the Sylvania machine.
The general construction of both readers is illustrated in the diagram. A fast-moving spot on the face of a circuit is directed optically onto the character to be read, and the reflected light is picked up by a photomultiplier tube.
In Philco's system three photomultipliers are being used so that a balanced-light signal is received from any point on the document being read.
A raster of 12 vertical lines is scanned over each character, and sensing takes place 22 times during each scan line. The transition from black-to-white or white-to-black is sensed. In order to eliminate possible false signals, from smudges or other noise, crispening is done in a video-processing section.
The electrical signal derived from the photomultipliers and subsequent La Placian correction circuits is in effect the sum of a twice-differentiated brightness signal in both the x and y directions. The resulting terms are then summed. The sign of the resultant is sensed to determine if a black-white or white-black transition is taking place.
The pattern of transition is fed simultaneously to a number of "masks," consisting of resistors of weighted values, depending on the importance of each point in recognition. If the summed output of the masking network is above a predetermined value, a trigger is activated, indicating that the character being scanned is the one represented by the mask being used. (Electronic Design, Oct. 11, 1961, p. 4)