Evaluating the new OCR desktop programs; today's optical character recognition software can read print better, and cheaper, than ever

Today's optical character recognition software can read print better, and cheaper, than ever

Optical character recognition (OCR) has come a long way since it could decipher only those funny-looking numbers the bank prints on your checks and the misshapen alphabets that went with them-called OCR A and OCR B.

Modern OCR equipment can read just about anything the human eye can read, except handwriting, and even that isn't far off: There's a prototype system in the Soviet Union, of all places, that can read handwritten script.

Like most other functions in publishing, OCR has now moved into the desktop world. Hardware and software for it have become powerful, versatile, accurate and fast, with prices to match the workload and complexity of the tasks it's called on to do.

For simple text, such as manuscripts turned out by a typewriter or a daisy-wheel printer, OCR software can be had for under $500-assuming you already have a graphic scanner. The combination of scanner and software can cost as little as $ 1,000. There is even a hand scanner that is sold with OCR software for only $199.

At the high end, if your workload is complex, you can spend up to $32,000. But between those extremes, a wide variety of OCR hardware and software is available for both the Apple Macintosh and the IBM PC and its clones.

Why do you need it?

The irony, of course, is that as OCR gets better and less expensive, the need for it in most publishing companies is decreasing. Probably the most common use of OCR in publishing is still reading manuscripts from outside contributors into a computer system, but a growing number of professional writers now submit their work by modem or disk.

However, it will be some time before all magazines can rely on receiving all copy on disk. Weaning writers from paper can be a slow process, as Cindy Rogers, executive editor for McGraw-Hill Healthcare Publications, attests. The Minneapolis-based healthcare group publishes four medical magazines, and the majority of its outside authors still submit copy on paper-"And once in a while, believe it or not, we still get a hand written manuscript," says Rogers.

The healthcare group is making it a priority to persuade authors to submit diskettes through such simple measures as promising to return the diskette. Still, its OCR system is bound to be around for a while yet: In February of this year, just 25 percent of the submissions came in on disk, leaving 412 pages to be scanned in.

In the end, even after all writers turn in electronic copy, there still will be many uses for OCR in publishing. Some publications that run large new-product sections, for example, scan press releases and then edit them. With sophisticated OCR systems, they can scan in the accompanying art at the same time. Other publications use OCR to scan tabular material from government releases and publications.

Choosing a system

If you're shopping for an OCR system, consider how you're going to use it. As the types of pages to be read become more complex, more software features are needed. For example, a magazine page with typeset text in three or four columns is a lot harder to decipher than a typed manuscript page, and it is more difficult yet if it includes headlines and artwork. Here are some of the key capabilities to look for when evaluating the latest generation of OCR software:

Omnifont capability. This means the system uses a technique called feature extraction to analyze the shape of a character in terms of lines and curves, rather than by comparing it against a few templates, to determine what letter it is. Size becomes irrelevant; anything that has the basic shape of an A will be identified.

Spell checking. Some programs include a spelling checker (which may have foreign-language dictionaries associated with it). Thus, "substitution errors"-reading a character as the wrong letter-will be caught automatically because the system sees a "word" it can't recognize. Some check spelling automatically during the recognition process.

Landscape mode. Pages that are wider than they are long, such as spreadsheet printouts or other tabular material, present their characters to tile system sideways. Many of today's OCR programs can read them.

Trainability. With this feature, the user can search for a character the program could not read and indicate the correct answer. The software adds the outline to its character table, and the next time the program sees that shape, it reads it right. This process can be time-consuming; high-end packages with omnifont capability don't need it.

Format retention. The more sophisticated packages will recognize such features as indents, boldface, italic or underscored characters, and save them in the file format of one of the popular word-processing programs or spreadsheets, selected by the user.

Style shoots. Programs that support style sheets enable you to scan documents into a standard text format regardless of their formatting on paper.

Page decomposition. If the program can read printed text on multicolumn pages, it must be able to separate the columns and read them into a disk file in the proper sequence. The more sophisticated programs can include any graphic images on the page, and display them next to the related text.

Deferred processing. Some packages can be instructed to scan in a batch of documents, but delay interpreting them until a more convenient time, such as the end of the day. If your scanner accepts an automatic document feeder, some packages will batch jobs that were batch-scanned earlier. If you place a blank page between documents in the stack, some programs will open a new disk file each time a blank page is encountered. These features eliminate more drudgery from the task of getting text from paper to disk.

Double-sided scanning. Documents printed on both sides of the page can be scanned to disk by packages with this feature. The stack of pages is scanned in with the first side up (preferably with the help of an automatic document feeder (ADF), then turned over and fed into the scanner again. The program then resorts the pages into their proper order.

These are some of the best of the packages available today:

For the PC

Omnipage 386. Caere Corp. $695. Most accurate and quite speedy, it has drivers for many scanners. With its ability to read from 6-point up to 72-point type, this package can handle everything from text to headlines.

WordScan Plus 1.0. Calera. $995. Easiest OCR program to use. Best at page decomposition, it reads the broadest range of scanned files but is not particularly fast. A Macintosh version was scheduled for shipment in April.

ReadRight 2.01. OCR Systems. $495. Comes in Windows and DOS versions. Requiring only 640K of memory, the DOS version is probably the fastest of all OCR software. The Windows version, It $100 more, needs 4 megabytes.

For the Mac

Omnipage Mac 2.12. Caere Corp. $695.

An automatic program, it works best with printed pages. Can scan both text and images well, but needs a separate pass for each. Requires OmniDraft to read dot-matrix files.

Accutext 2.0. Xerox. $799. AccuText recognizes words and characters by shape and content, using its dictionary to understand fuzzy or otherwise difficult images. In a single pass, it can scan a page with both text and graphics.

Alfred D. Berger is a veteran magazine editor who has been running editorial computers for more than 10 years. As manager of editorial systems at McGraw-Hill, he selected and set up his first OCR scanner in 1982.

COPYRIGHT 1991 Copyright by Media Central Inc., A PRIMEDIA Company. All rights reserved.
COPYRIGHT 2004 Gale Group