There’s some piece of freeware that I’ve lost (mostly I search the pirate sites for scans, as its easier, than scanning in myself) which will clean up scans and so improve OCR. Its part of the Linux OCR suite that uses Tesseract (which is okay for free, but if time is money…), and is very useful.

Incidentally, a LOT of academic books are out there on filesharing sites these days. I’m not sure what that’s an indicator of.

• Which of these programs is good at recognizing when symbols don’t make sense to it (e.g. an equation) and storing the resultant information in a picture, and just including that picture in the resultant file?

• Which of these programs does things like recognize word frequency in order to improve its ability to OCR correctly in ambiguous cases? (Or, alternatively, can be trained to prefer certain words over others.)

• Which of these programs can be “trained” to recognize new symbols and substitute them, for example seeing ║ and OCR’ing it as norm, or seeing β and OCR’ing it as beta ?

• Which of these programs can be trained to recognize formatting data and encode that: for example, scan “what dreams may come” and OCR this as “what begin{boldface}dreamsend{boldface} may come” ?

• Which of these programs can be trained to recognize relative position, for example notice e^x and OCR this as esuperscript{x} ?

I am planning to achieve world domination by developing an open-source OCR tool that scans old math texts and converts them to .TeX files which, when compiled, faithfully recover the original text and formatting, but as I currently have exactly none of the skill set necessary to accomplish this, I’d rather get the next closest OCR program created by someone else and tinker with it a bit.

Thanks in advance to anyone who responds!

With Acrobat, I found it best to use the option to retain the bitmap text, rather than to replace the bitmap with the OCR’d text. The text is searchable, and copyable, but the original font remains, and the OCR mistakes are hidden. (Which could be a problem, but might not be.)

The tools are somewhat expensive, but once the investment is made, conversion from deadtree to e is fast … assuming a recent printing with clear, crisp print.

Be prepared to break its spine. As to sheet feeding: I’m sure you don’t want to be feeding actual book pages through one of those, even if the whole spine has dissolved and you just have a pile of paper. You could photocopy the whole, fairly quickly, then sheetfeed the photocopies, more slowly. That might produce perfectly good results. I have access to a really high quality flatbed scanner at work and a less high quality flatbed at home. I think flatbed is probably the way to go with old books.

I picked up an old scanner with a sheet feeder at a ‘freecycling’ event here at work. Had to buy a semi-compatible power supply, and need to solder it in, but it works, more or less.

Some of my books are starting to turn yellow, so retaining the physical objects isn’t a strong motivator.

Fortunately, Acrobat did a pretty serviceable job on the thing Quiggin sent me. Everyone with Acrobat should know this: there’s an OCR option under the document menu. (No great shakes, but it works.)

John Quiggin: Send me that PDF you want converted. I’ll send you the RTF conversion that is produced with 1-click. And you see whether you like the results.