Can't the OCR program be taught to translate certain shapes and images to certain characters? So if you tell it once that 2.AN. is a W., won't it know from then on? Or is the method you're using not that interactive?
Being a programmer myself, I'm really interested in how exactly you're doing this. If there's something I can do to help, let me know.
Otherwise, I'll be waiting to hear more on whether you can perfect the OCR. Even if you don't manage to perfect it, correcting a few faults in the OCR result by hand and then pasting into NST would still take much less time than actually entering everything on your own. So even if you only achieve a near-perfect result, I guess we could live with it.
EDIT: Just saw your edit; I had no idea Python made OCR THAT easy and straight-forward. I've learned and used Python as part of my Computer Engineering courses, but never tried OCR with it. Is there a way we could feed Pytesser some corrections and have it learn from them, kind of like OCR suites do? You've got me curious, I'll have to look into it tomorrow and help you with this, unless you manage to perfect it by then.
Sorry I can't stay for now, it's 4.30am here, and I'm sitting an exam tomorrow. Good luck, and I'll try and help after the exam tomorrow.
EDIT 2: A quick Google search led me to a StackOverflow question which recommends
Tesseract as an even better (and still free) OCR solution. See if you can get better results with it, or I'll test it tomorrow.
Here's the relevant question:
http://stackoverflow.com/questions/5925 ... inaccurate