Do you really Caere about OCR?

Optical character recognition used to be expensive and none too good, but about 3 or 4 years ago Caere's OmniPage program tipped the scale enough that it became faster to scan text and clean it up than to retype it from hard copy.

Version 8 of the software was the first version I could recommend without reservation. Then version 9 raised the standard. Now version 10 has gone beyond that standard. I won't say it's perfect because Caere is probably working on version 11 - but it's very, very good.

When you need to edit or reuse a document that you've received on paper, you have two choices: Type the material yourself or use an OCR program. No matter how good the OCR program is, it will invariably make mistakes and you'll have to fix them. Version 10 of OmniPage Pro is surprisingly accurate even with less-than-perfect copies and dead-on with good copies.

The program runs under Windows 9x, Windows NT 4, and Windows 2000. New to this version is a voice read-back capability that enables users to verify OCR results. It could also be used by sight-impaired people to read printed materials.

OmniPage Web (Personal Edition) is in the box with OmniPage Pro, making it possible to convert multi-page paper documents into hyperlinked Web sites. The program tries to recognize the hierarchy of the paper documents and create links accordingly.

OCR used to be a complex operation. You had to fiddle with the scanner to get it in the proper mode. Then you had to scan the page. Next came manually marking columns of text. Then the recognition phase. And the correction phase, during which most users decided that retyping the material would have been easier.

That's not the case with OmniPage Pro. If your scanner has a page feeder, you can scan multiple pages by pressing a single button. I still prefer some human intervention, but the program is surprisingly accurate in automatic mode - determining columns, text flow, graphics, headlines, and such.

OCR parallels speech recognition

Both optical character recognition and speech recognition have a long history in computing. Both were envisioned at a time when computers weren't powerful enough to do the job - not even the old "mainframe" computers.

The earliest OCR program was introduced in 1959 by the Intelligent Machine Corporation. It could read just one font in one point size, and was used for processing preprinted mortgage loan applications in the banking industry. Later, programs that could read nearly a dozen typefaces were developed, but they were accurate only when the operator selected the right typeface library.

In 1966, an American standardized font called OCR-A and a European font called OCR-B were developed. You've probably seen these in your list of fonts. This is OCR-A.

Kurzweil Computer Products introduced a system in 1978 that could be trained to read any font, but each new typeface required several hours of training time.

Until the early 1980s, OCR systems were rule-based. They broke each character image into a set of lines and curves and then determined which character most closely matched the extracted features. This method worked well as long as the original was clean.

In 1986, Palantir introduced an "omnifont" that could read many typefaces and, instead of being rule-based, used new technology - "neural networks". This is a technique that allows the computer to learn. Later, Palantir became Calera and in 1993 the company introduced "Adaptive Recognition Technology" that improved recognition again.

Caere Corporation developed Language Analyst software that added linguistic information and the ability to examine three letters at a time to look for common patterns. Caere also added a dictionary to help improve accuracy.

OCR programs began to use a variety of tricks to figure out what they were looking at. They began with letter forms, but they also considered spelling rules. Caere refers to these various technologies as "experts" and during recognition, each expert had a vote. The interpretation that receives the most votes is the one the program selected.

After purchasing Calera, Caere combined technologies developed by both companies. When the individual characters in a word were difficult to isolate and recognize, as was common on degraded document images, "Predictive Optical Word Recognition".

This technology has been tweaked, tuned, and tinkered with over the past few years. Computers have become faster and more powerful. Combined, these forces make it possible for version 10 to be faster and more accurate than any previous version.

Do you need OCR?

Not everybody needs OCR, but if you're someone who does, make sure you see OmniPage Pro before you buy any other program.

How much should you pay? $99 You'll probably see a higher price because $99 is the "upgrade". This is the price you should pay even if you think you don't own any OCR software. If you have a scanner (if you want to do OCR, you need a scanner!) you probably have OCR software. Most scanners ship with some OCR product. That's enough to qualify you for the upgrade.

Don't pay more than you have to.

For more information, see http://www.caere.com/.

   
 
 

William Blinn Communications - All Rights Reserved
Subscribe to the Newsletter
179 Caren Avenue • Worthington, Ohio 43085
Phone: 614-785-9359
(Telemarketers please note: This number is listed on the national Do Not Call registry.)