A district supervisor decides to update policy manuals created five years ago, before computers were integrated into his office. He, or rather his secretary, faces hours of typing in order to make either a few or numerous corrections to the old text. Hours of productivity are lost.
An English professor wants to incorporate several short stories into her curriculum materials, which are all organized on her personal computer. Her only option appears to be to use the department's photocopier, whose output is often poor in quality, and generate handouts to go alongside her materials. But adding her own notes and coordinating the texts with the course's theme are difficult and the end product lacks cohesion.
For these or any other educators and administrators who wish for a way to magically transform a document into a computer file, there is an answer--optical character recognition (OCR). Achieved via software and a scanner, OCR is the missing link between hard copy and a word processor, allowing users to import documents; edit, change and delete copy; and print out final versions or integrate the material into an existing electronic file.
There are several pieces to the OCR puzzle. Needed are a computer, a scanner (flatbed, sheet-fed or hand-hold), OCR software and a word processor. All of these pieces must also be coordinated--although many scanners are compatible with either DOS or Macintosh computers, OCR software is much more specific. Versions for the Mac, for instance, generally require a Mac II computer with 2MB or 4MB of available RAM memory.
New on the market this year are numerous packages offering Windows 3.0 support and the first OCR program for the Apple II family--InWords by WestCode Software.
Also, the OCR software needs to be compatible with both the scanner and the word processor being used. (The latter is rarely a problem, however, since the majority of OCR packages save files in ASCII, a universally accepted file format.) When all of these pieces fit together, the end result is a coordinated OCR system.
* The Hows and Whys
The process is as follows: First the document is fed through a scanner, which optically reads the page at a given resolution (usually 300 dpi). A large grid is formed, and each cell in it contains a pixel which is either on or off, represented by either a black dot or white space. The black dots then become individual characters.
A bitmap or scanned pixel image of the entire page is stored in the computer's memory to be accessed by the OCR software. The software identifies areas of black as characters, each of which is then compared to type tables that are loaded into the computer's memory. When a match is found, the corresponding ASCII character is entered into a text file.
Some packages are billed as "trainable" OCR programs. This refers to their ability to be programmed to recognize font types not found in the program's table. Trainable programs instruct the user to type in the correct letter that matches the unknown character. Once established, this new font style can be saved in the table and called upon when needed.
OCR packages are not infallible; errors appear when characters are unrecognizable or no exact match is found. A substitution error occurs when the best match chosen by the software is not correct. Common substitution errors are an "m" for "rn, "n" for "ri" and "x" for "oc." Often no match can be divined by the program, so to indicate its confusion, it replaces the unknown character with a dingbat item such as a tilde ( ) or an asterisk (*). There recognition errors often result from scanning poor originals such as those produced by faint dotmatrix printers whose characters are not solid.
Who's Doing What
The most profound aspects of OCR are the speed with which the technology is changing and the rate at which it is also becoming more affordable. …