grafix I: AN IMAGE PROCESSOR THAT READS HANDPRINTING…

摘要

Machine reading of hand printed letters and numbers has been the goal of several organizations for many years. There has been some success in reading machine printed material, but the reading of hand printed material was considered too difficult except for the limited range of zero through nine and a few letters. Now, a significant advancement has been achieved by Information International Inc. (III) of Culver City, CA. The entire upper case alphabet, numbers zero through nine, and some punctuation marks can be read with high accuracy and little training of the people who do the writing. Performing this task requires a new combination of hardware and programming techniques. Much research was done in order to design this system. The results of this research was implemented partly in the hardware, and partly in the soft-ware. Shown in Fig. 1 is the block diagram of the entire system, called GRAFIX I. Heart of the system's computation power is the Binary Image Processor (BIP), a general purpose image processor designed by III. Probably the most difficult concept to grasp about this system is the fact that the BIP is a general purpose computer oriented toward high-speed image processing computations. At first glance, because the BIP is only involved with image processing, it appears to fall into the general category of a “dedicated computer.” This is not so because the BIP was designed to do a wide variety of jobs within the general area of image processing. More about the BIP later. Additional major components of the system are a 36-bit general-purpose central processor, large core memory, a precision CRT film scanner, and standard console interactive operator computer peripherals. All material to be scanned must first be filmed. The film is mounted in the GRAFIX 1 film scanner (a special design by III), so that processing may begin (Fig. 2). In processing an individual handprinted (or other) character, three results are possible. A character may be identified correctly; it may be incorrectly identified (this type of error is called a “substitution”); or it may be “rejected”, or not recognized at all. In the case of a rejected character, a stored picture of it is displayed on a CRT, together with its line of text for identification by an operator (Fig. 3). The BIP (Fig. 4) is the unique piece of hardware that allows the GRAFIX 1 system to achieve both extreme flexibility and high throughput speed. Built of medium-scale integrated circuits, the BIP contains a large amount of pipeline logic running synchronously at 40 MHz. The BIP is unlike any general-purpose computer in its organization. As a general-purpose image processor, some of the tasks that can be performed are: conversion of image gray scale data into binary data; binary image enhancement; measurement of geometric image properties; image transformation or rotation; changing of image size; and comparison of an unknown binary image with a reference binary image. High speed efficient operation in this serial processor is due to several facts. Primary among them is that in the BIP, loops are achieved through hardware, not software. Large amounts of data are handled in each instruction. For example, in most conventional general purpose computers, it takes many instructions to count the number of ones in a binary image (part of one recognition technique); the BIP has built-in circuits which accomplish this task in the execution of just one instruction. Original goals in designing the BIP were to create an image processor which would be both general enough to apply to new applications without hardware redesign, yet compact enough to be economically manufactured. The original design served was 1968-1970. Normally one would expect a first-of-its kind design of that period to be obsolescent by now. However, using it over a period of five years and on a number of different applications, its design goals have been met, and there are a large number of new applications suitable for the