by Dana Trout
March 5, 1996
This Optical Character Recognition (OCR) program is quite an advance over the earlier ones.
I've used earlier versions of TextBridge for years and have been generally happy with it for OCRing simple text pages. The interface was clunky but the recognition was quite good and the price very attractive.
This new version has added features which make the program more generally useful: it can deal with multicolumn text, embedded graphics, automatically correct for upside-down or sideways (landscape) scans, and be used from within popular wordprocessing programs.
The installation is straight-forward (especially if you remember what port or SCSI address your scanner uses) and the default settings are reasonable so you can do a sample OCR immediately after installation.
Although the buttons and drop-down menus are not labeled, if you leave the mouse cursor on any item for a second or so a yellow box with descriptive text appears (this feature is sometimes called "balloon help"). It is a reasonable way of limiting the amount of screen space devoted to icons and labels while still giving the user the needed information.
This version of TextBridge Pro (TBP) is much more usable than previous versions of TextBridge (TB), primarily due to the addition of one little button: the Cancel Current Page button. The reason this button is so useful (and a major advance from TB) is that if you are scanning a multipage document and one page is upside down, or lighter or darker than usual, or just didn't feed right, you can cancel just that one page, take corrective action (rescan right side up, adjust the brightness and rescan, whatever), and continue scanning and OCRing the rest of the document. In contrast, the previous versions of TB required that you end the current session on a problem page and start a new session, with the result that you had to mess around putting together multiple jobs as one document when you were finished.
The buttons, as shown above, give you immediate access to the most used preferences and functions. The GO button tells TB to perform the next step, such as initiate a page scan. You can set the preferences to be as interactive or automatic as you wish: you can tell TBP to consider one step to be as little as scan and display a page to as much as scan and OCR all pages in the scanner (this assumes you have an automatic document feeder (ADF)).
The STOP button tells TBP to abort the current step. The Cancel Current Page lets you cancel a bad page scan, make adjustments, and then rescan without having a superfluous page image.
The next two buttons let you select the source of the page images: either a scanner or from disk files.
The last three buttons are on/off: Page Preview will cause TBP to stop after each page is scanned and display its image on the screen. This is handy if you need to check the scanner's brightness setting or if you need to select text blocks for each page. You can turn training on or off with the Training button. The last button is Save Page Images - Defer OCR. I'll explain what it does when I go through the Process menu.
If all your pages have the same layout and print density the simplest way to use TBP is to turn on Preview, do a sample scan of one page, and set the scanner's brightness, the page orientation, document quality (fax, dot matrix, normal), and select the area(s) to be OCRed. Once these settings are made, then you can turn off Preview and let TBP read the pages and OCR them by itself while you go off and do something more entertaining.
TBP has the capability of automatically determining the proper document quality, orientation, and can deskew page images. But these features all come with a price: slower throughput. If you are doing more than just a page or two, it pays to pick the appropriate settings manually and to make sure the pages feed through the scanner with little skew.
TBP has several interesting features that show up in the Process menu shown above. One, of course, is the ability to see a page preview after the scan and before the OCR. Another feature lets you set the training level -- it lets you tell TBP whether it needs lots of examples or just a few. You would choose Fewest if the characters are very consistent (as from an original laser-printed document), while Most is appropriate when there is great variation in the characters (as from an nth-generation copy). The Save Page Images - Defer OCR feature is important because scanning can take a significant amount of time, so TBP lets you scan up to 50 pages at a whack without doing the OCR right then. This is especially useful if you have a scanner with an Automatic Document Feeder (ADF). After the pages have been scanned you can sit down and OCR the scanned pages interactively, selecting the proper areas and correcting recognition errors. It's nice that TBP provides this feature and it's nice that it lets you choose TIFF as the file type, because a lot of programs understand TIFF. But I do wish that they would let you choose a more compact file type as well.
The Save Proofing Data can be very handy. What it does is it keeps a snippet of the page image that shows each questionable word, so when you use the TextBridge Proofreader in WordPerfect 6.1 or Microsoft Word 6.0 and you come across a questionable word you can show the image and see what the scanner saw. It sure beats having to go back to the original paper document!
You noticed that I selected Preview so after a page is scanned I can look at the image, as shown above. Of course what I've shown here is too small to check the quality of the text scan (you want the brightness set dark enough so the characters are not broken, but not so dark that the characters fill in or touch one another). There are several important buttons that appear in the page preview: the Zoom In and Zoom Out, the Text Zone (the big T) which lets you draw rectangular frames around areas of text you want to OCR, and the rotate buttons (90<198> ccw, 180<198>, and 90<198> cw). The one thing that annoys me here are that the zoom in and zoom out are two separate buttons. That means that if you have zoomed in and now want to zoom out you have to move the mouse to the toolbar, choose the other zoom tool, then click on the page image. I really prefer being allowed to use Ctrl, Shift, or Alt to select the zoom out -- it's much faster. Also, the zoom in tool does not allow you to draw a marquee box around the area you want to see in greater detail -- instead you must click in the middle of the area you want expanded and the image is enlarged to 200%. Another thing that annoys me in the preview window is that even though TBP lets you select multiple text blocks, unlike TBP's competitors which also allow you to reorder the text blocks, TBP does not -- you have to start all over again. If you never make a mistake this is not a problem.
You can see from the zoomed-in preview that the scanner brightness is set too light -- many of the characters are broken. I did this on purpose so I could show how well the TBP Proofreader works: if there were no errors there would be precious little to show!
This screenshot was taken in WP 6.1, and shows TBP has installed its proofreader in the File menu. When I clicked on it it brought up the dialog box shown below (TextBridge In-Place Proofreader) which stopped on the suspect word "JEOPARDY!" I then clicked on the Show Image button and got the following display: the image is a bitmap image of the original page and shows whatever was there. Very handy, especially if there are handwritten notes (which do not successfully OCR at all).
If you look closely at screenshot above you will notice the second line of body text starts out "1)isabilities Services..." In other words, TBP recognized the broken D as "1)". Now TBP uses color coding to indicate its confidence in recognizing the words: black text means TBP is absolutely certain of its recognition, green is high confidence (but not certain), yellow is medium confidence, and red means the word (or letters) are highly suspect. And if TBP couldn't figure it out at all, it puts in a tilde (~). Note that the characters "1)" are black: TBP was absolutely certain that the recognition is correct. The point I want to make is that you still need to proofread an OCRed document even if the recognition seemed to be fine. Just because a computer says it's so doesn't make it so.
This screenshot shows the Show Image function.
This image shows the dialog box that shows up while you are training TBP. It stops on each word it is unsure of and asks for confirmation or correction of the recognition. From your responses TBP "learns" the shapes of the characters. In this case the scan of the word Disabilities was pretty crummy (look at it in the Image window) so TBP is uncertain that it recognized the word correctly.
The proofreader in TBP is helpful, but would be even more so if they chose a different font. A common error in any OCR program is to recognize an "m" as "rn" or vice versa -- with the font Xerox chose for the proofreader it is hard to tell if you are looking at an "m" or a "rn".
Another operational issue is that sometimes you need to adjust the brightness setting. The problem is that you are given only three choices: darker, normal, lighter.
Xerox TextBridge Pro ver 3.02
Street price approx $250