Slashdot Mirror
Best OCR for Technical Texts?
An anonymous reader asks: "I'm scanning in user manuals for older lab equipment. I've never used OCR before today, so I installed the Caere Omnipage 9.0 that came with the scanner. I was pretty happy except for a few things. It doesn't seem to want to recognize engineering symbols like the one char +/-,square root, omega, simple equations, it has trouble with super- and subscripts, and it outputs funky Word files. For example, from an 8.5 x 11 original page scanned in at 1 bit at 300 dpi, the output Word file was 10 inches wide, used tons of Omnipage text styles and didn't match the original text's flow. It did do a good job of italicizing headers and recognizing the various sections in a two column page. Googling the news and net just backs up my claims but provides no real solution. A Google search that provides nothing useful looking for best OCR for engineering."
Have you looked at the open-source Clara OCR? I've used it for some very unique texts in the recent past. It's accuracy is quite good. Besides that, the proofing mechanisms are great!
Go here: http://www.claraocr.org/.
It has very recently been ported to win32, and the community support (via e-mail lists) is excellent.
That might help slightly...
Good luck!
I've used a few different version of Omnipage PRO, and it works OK if the layout is not complicated, it uses standard fonts, the text is clean and clear and it doesn't have too many weird logos or symbols. You still have to proofread everything and correct it by hand, though, so I'm not convinced it's a time saver as much as it is a typing saver.
OmniPage Pro does do a MUCH better job of identifying words that the free version they throw in with scanners because it uses spelling and grammar checkers to help ID words from context. The free version is as close to useless as you can get in the software world - it's really just an ad for Pro.
Engineering and math symbols are right out.
"Lawyers are for sucks."
- Doug McKenzie
Have you tried other combinations of settings (e.g. dpi, bit depth)? That won't solve all of the problems you talk about it, but playing with those settings in each package you look at _before_ rating how good it is is important.
the clock on the wall says 4 til 7
You can try FineReader from ABBYY
Use 8 bit, NOT 1 bit. When I switched from 1 to 8 bit on a page of normal text, the dozen or so errors vanished.
: //docmorph.nlm.nih.gov/docmorph/
l l. asp?category=ocr4
s ource. htm
Since Omnipage is up to version 12, perhaps there's been an improvement since your version.
Your google skills are sorely lacking, the "Hacking Google" book would be a good investment for you. Eliminating the quotes and word "best" in your search string would help.
2 different free web based ocr, just upload a 300 dpi b/w (8bit greyscale) file
http://www.expervision.com/webtr6.htm
http
here are some OCR programs
http://www.scansoft.com/omnipage/
http://www.abbyy.com/
http://www.newsoftinc.com/redir/digitaloffice_a
more ocr links than you really want
http://web3.humboldt1.com/~jiva/ocr/_ocr_re
www.expervision.com/webtr6.htm
http://docmorph.nlm.nih.gov/docmorph/ here are some OCR programs
http://www.scansoft.com/omnipage/
http://www.abbyy.com/
more ocr links than you really want http://web3.humboldt1.com/~jiva/ocr/_ocr_resource. htm
I can't speak for the rest of the programs you mention, but Abbyy doesn't recognize equations very well at all. (Based on the last time I tried which was about a year ago).
What you really need is ICR, Intelligent Character Recognition. There is a free trial version of one such product here.
Better Google searching makes the difference.
Operator, give me the number for 911!
Anyone know of a decent home-use scanner with a letter-size sheetfeeder?
Can it also take a stack of 4x5 photos?
The Best OCR scanner is an intern with a pencil. ;-)
Good quote, too many chars. Seriously, the slashdot 120 char limit sucks!
Oh boy, it's been a long time, but when I evaluated OCR packages I found that each of the 3 major choices at the (OmniPage, TypeReader, and one whose name I no longer remember and is no longer available) could do better than the others on certain types of my test documents. So you might want to get a trial of TypeReader (www.expervision.com) and try it out, in addition to FineReader as suggested by another poster.
Abby Fine Reader
I got a lot of interesting results Googling for "ocr superscripts symbols".
Here's my (non-copyrighted) strategy for doing a Google search. Google is fiendishly fast (which I find mind-boggling, given the size of the database!), so there's no reason not to play around. Start with an absolute minimum of keywords. If your results are too broad, add one or two keywords and search again. Iterate until you have useful results or you reach a dead end. If you do reach a dead end, the browser's "back" button is a convenient way to back out to a broader search.
I find the Google Toolbar indispensible. It has a lot of features, but only three that I ever use:
- A handy search text/list box. Not only does this it save steps while entering a search string, it automatically syncs itself with any Google search you enter, even if you do it just by back-buttoning out to a previous Google page.
- A "search this site only" button.
- Automatically generated buttons that search the current page for your search terms. These are real time-and-aggravation savers on a lengthy search.
I also use the uplevel button, but that's really a patch for a missing Internet Explorer feature.If you're a die-hard Netscape/Mozilla person, there's a Sidebar with most of these features. Notably missing are the automatic term buttons -- main reason I still use Internet Explorer.
I've found that the best is to leave the scans as image files and bundle them together as a multipage TIFF or PDF file. It takes more space to store them, but you don't have to mess around with OCG (Optical Character Guessing).
Easy to access and read. The only loss is you can't do cut and paste or text searching.
Then, use something like Adobe Acrobat to put them on-line: Acrobat uses OCR internally to make the text searchable, but it still displays the original page image. That means that formulas and appearance will be preserved even if the OCR screws up.
the newest version is much improved.
I have not tried to use it on technical articles, but it is very good at scanning normal documents and keeping both the fonts and styles intact.
http://www.totalsol.com/products/doc_process/prime recognition/product_prime.html
How about Gamera Gamera: A Python-based Toolkit for Structured Document Recognition http://dkc.mse.jhu.edu/gamera/papers/gamera_python _2002/gamera.html