NWA-PCUG Newsletter Article, November 2002
OmniPage Pro 11
A Product Review by Karl Rehak, LVPCUG
karl@idea4u.com
(click to email author)


Optical Character Recognition (OCR) software has been around for many years. Anyone who has had a need to capture an existing document for revision knows the value in having this kind of product to remove much of the burden.

A little background of my own experience with OCR software might be in order before beginning the review. I have been a user of this kind of software product for over five years. Starting with the light versions of OCR products with my first scanner followed by two releases of TextBridge. This my second experience with OmniPage Pro. The first OmniPage product I used was release 10.

My general opinion of OCR software is that while it removes most of the effort in capturing printed documents, the OCR process is in itself a very burdensome activity. The software is only capable of doing so much of the effort and any ambiguities in the OCR process requires interpretation that only a human being is capable of providing. As a result I haven't approached an OCR session with delight or as being an activity that would be personally rewarding. Additionally, my OCR sessions are generally infrequent and usually come in bursts that can range from several hours to several days. This means that ease of use is important and the more accurate an OCR program is, the easier will be my brief but intense excursions into their use.

OmniPage Pro release 11 is the first release since Caere was bought and ScanSoft merged their OmniPage OCR program into the TextBridge product. This is also the first new release from ScanSoft in just under two years.

Needless to say, and based on my past use of OCR products, I was very interested to see what has been brought to the scene. First and foremost is the claim that the product is more accurate than preceding products and my first question - what does that really mean to me? A feature called IntelliTrain has been enhanced and once again the question - what does that mean to me? Again, because of the intense interaction during OCR sessions, those were the two most significant areas of evaluation to me. Several other new and enhanced features were evaluated as the product was tested and comments regarding their value will be discussed through the review.

The bottom line measure of the value of a quality OCR product is end user productivity. As the quality of OCR products improves and the interpretation of a document's content becomes more and more a function of the computer's ability to interpret information rather than the time spent to have a human being interpret and make decisions, the higher the productivity. That is, in the final analysis, all that matters. So along the way judgments of the OmniPage improvements on user productivity will be highlighted.

The Test Begins
The first test was done primarily to allow me to gain knowledge and confidence in the use of OmniPage 11. I read the instruction manual and reviewer's notes and formulated a test plan to put the product through its paces, but first I had to know what capability was in the software. For this first test I used the documents out of the OmniPage box. The documents are very legible and of a standard size font, about 12 point Ariel. I rated each test based on the overall quality and legibility of the source document, a 10 is a very easy document and a 1 is an extremely poor quality document. These first test documents were quality 10.

The overall design of the main OmniPage Pro OCR process is built upon the age-old concept of what computing really is. The steps are Input, Process and Output - a simple concept. Add into that each of those processes can be broadly segregated into three different approaches. They can be done manually, automatically or described through a wizard. I tried each method as I tested and found that they are very similar and easy to learn and use. It takes no additional effort to move from one approach to another.

The input describes the source of the OCR, it can be a black and white, grayscale or color document. It can also be a computer file. Introduced with this release is a scanable file type, Acrobat files. More on the PDF scanning ability will be covered later.

The process part of the three elements allows the user to describe what he or she wants as a result. Among the choices are to preserve the page structure of the source, scan it as a single continuous column of information, include or exclude tables or to output to a spreadsheet format. The program also recognizes over 100 different languages.

Finally, the output function allows the user to save the information as a file for a word processor or other computer program, output to mail or copy the output to the clipboard. There was another interesting output available speech. More on that later.

The first scans from the manuals went as expected. It interpreted everything with only three suspect words in three pages of scanning. A word with a registration mark (the circle R) next to it is a example of a suspect word. It guessed correctly and I only had to confirm the program's choice. The interpret rating was 100% and again, the document type was a 10.

Test Two
Alright the parties over, time to see just what this program can do. For second test I selected a sampling of a legal document, reproduced many many times, originally bound in a plastic binder with half inch slots, later converted to three hole punched paper. The document because of numerous copies over the years was heavily speckled. I called it the document from hell. This is characteristic of many of the scans I had done in the past and where the personal productivity factor went in the dumper. Lots of interpretation needed and in the older OCR programs the remnants of holes and slots were interpreted as characters.

The documents also had to be cleaned of speckles and marks manually with older OCR programs. Further the document had a filing number stamped on it of a different size and slant. Document quality 2 at best on my scale,

I decided to name this part of the test 'Extreme Scanning.' The illustration, "Legal Document," shows an image of the document quality. Four pages were scanned and the overall scan accuracy was 93.93%. On the first page it highlighted 17 words. Of the selections it had guessed correctly 15 times and I had only to confirm the choice. I had to select from a provided list of choices twice, and the correct words were in the list.

I'm convinced that lawyers invent words so nobody knows what they are talking about. The net in the OCR test was that the document was loaded with legalese and OmniPage Pro flew through the legal words just fine.

The additional pages had slightly less selections to decide, apparently the IntelliScan feature was doing its job. The parts of the documents that contained remnants of holes and slots were passed over by the program, no need to clean them beforehand or eliminate wrong interpretations on the part of OmniPage. Its beginning to get too easy for this program (and me), so for the final scan of the legal document I decided to place it in the scanner at about a 10 degree angle. It scanned fine. The program automatically aligns the text. When I took the document off of the scanner I realized I had also put it in upside down. The program had also fixed that. I became a believer.

At this point I was sold, the effect on my overall productivity in the OCR area was great. No more speckle and hole removal or spending countless hours interpreting information. But I figured there had to be one more test I could do.

Test Three
I had achieved my goal of watching the program pass the extreme scanning test, but what else would put it to the test after that? Looking over my bookshelf I saw an old copy of the Bible. The pages are 4 « inches wide and 7 « inches long, the font was too small for me to measure but is about 7 point or less. Lots of unusual words, numbered verses and italicized footnotes. I placed it on the scanner at a 10-degree angle and hit the scan button. The illustration, "Interpreting the Bible" shows what the program saw. The result was 100% accuracy, 13 suspect words and no changes. On the scale of document quality I put it at about a 6.

I'm amazed at this performance and now we're in bonus territory in our extreme scanning test.

The Bonus Round
Added to the program with this release is the ability to OCR Acrobat files. To do this test I found areadme' file in my Acrobat folder and scanned it in. It is a 21-page file and all 11,000+ words were input at 100% accuracy with 64 suspect words. Again, the registration symbol and similar characters required confirmation that the interpretation was correct. No changes had to be made.

Added to this release is voice output. I had a lot of fun with this. There are a variety of voice types but the one that was easiest for me to hear was called "Sam." Sam sounds like Steven Hawking. What is the value of this you ask? If you hold the original in your hands, the voice will read what it has and you can do a word-by-word review of captured information. Sam will not have much of a career as a stand up comic, but he can come in handy. Personally, I doubt that Sam will have a serious place in my OCR activity, when the OCR process is done I have a very high level of confidence that it is correct and the voice read back is a redundant activity. Perhaps future releases will find more for Sam to do.

The program supports over 100 languages however, Chinese, Japanese and Hebrew are not among them.

Conclusion
All in all ScanSoft has delivered a product matched to its promise. The product is more accurate than previous releases and is easy to use. As each release improves accuracy, the user productivity soars, I would estimate that productivity improvements of an order of magnitude over previous releases can be realized. What used to take 10 hours of effort can now be done in 1 hour. The more complex the document and the poorer the condition the more work the current program will take on, relieving much of the effort of capturing documents.

The program test was a lot of fun to do. The burden of Optical Character Recognition has been lightened. ScanSoft has delivered a significant release and OmniPage Pro release 11 is its name.

Click here to return to top



==================================================================