[PLUG] OCR software?

[John Jason Jordan]

On Wed, 5 Apr 2006 09:17:55 -0700
Eric Wilhelm <scratchcomputing at gmail.com> dijo:

> I'm guessing that you might want to scan the page to a (pbm) file and 
> work from that while you're trying to get the OCR working just-so.  If 
> xsane runs the scanner every time you try this, that extended wait is 
> going to be frustrating.

The problem is that Xsane -- as cool as it is -- seems to have no
ability to OCR from a PBM file (whatever that is -- something like a
TIFF?). I can save to a PBM file from Xsane. It's just that a PBM file
is all that it will remain.

> What resolution is your capture?  200-300 dpi B&W is typically 
> considered to be the minimum.

Well, I can change that in Xsane, and I tried it, but I still got lousy
OCR results. The problem is that there is no ability to tweak the OCR
end of the process.

> >The problem is I can't figure out how to tweak gocr. From my
> > experience with OCR programs on Windows there should be a "learning"
> > tool and other goodies to enhance the results.
> 
> Were those goodies free?  GPL?  Then you could just port the relevant 
> bits to linux and be all set :-)

Actually, yes they were free. On the Windows desktop computer I can use
any of several competing brands under a "free trial" period. I have yet
to buy an OCR program for the Windows machine. But then, I don't need
to do it very often. As for porting, sorry. Not a programmer.

> OCR is not an easy problem.
> 
> http://kooka.kde.org/ is only going to be a different front-end and will 
> either interface with gocr, ocrad, or a commercial app (if you want to 
> spend $1k on it -- sounds like a $100 donation to gocr or ocrad might 
> be a good investment :-/

OK, I found kooka in Synaptic and installed it. Much better! Now I'm
getting somewhere. See, I figured gocr or ocrad was somehow tweakable.
They had to be, it was just inconceivable to me that Linux OCR software
was as bad as the results I was getting. It's just that the Xsane front
end doesn't implement the buttons to do so. Kooka does. So now I am
working on figuring out kooka.

Only problem is, now I have installed yet one more KDE app in my Gnome
desktop. I have so many KDE apps installed I might as well be running
KDE. Except I still think KDE is ugly. :(  Oh well. You mentioned that
the OCR process was slow. Using kooka I did a preview scan, selected
one long paragraph from a really clean page (no stray marks or junk on
it), and then told kooka to OCR it. The gear-thingy started jumping
around, indicating it was working on it. Five minutes later I killed
the process because something must have gone wrong. Surely it can't
take five minutes for one paragraph (about the length of this
paragraph). I can type way faster than that. Methinks it got hung up
somehow. I tried it several times and got the same results.

Anyway, thanks for the suggestion to try kooka. That's a big
improvement. Now I just need to figure out what is wrong with it.