[Bug-ocrad] Possibility of dictionary-enhanced OCR in Ocrad

[Martin C. Doege]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Antonio!

First of all, thanks for taking the time to develop Ocrad! I 
particularly like how incredibly fast it is in comparison to, say, 
GOCR. And the recognition rate is not too shabby for a character-based 
OCR program.

That being said, I do wonder if it would be possible to extend Ocrad 
with dictionary lookups to improve its recognition rate. So for 
example, it would work on a word and assign each character a confidence 
value. So it might recognize the word "heflo", and since that word is 
not in the dictionary, it would tweak the characters with the lowest 
confidence value first.

So in this example, maybe it would happen to have a low confidence 
value for the "h" and the "l" and therefore try to find permutations in 
the dictionary: "beflo", "bello", ""hello",... If this is not 
successful it could try replacements like "m" -> "ch", "l_" -> "n", 
"ii" -> "ü", or whatever other common errors are found in OCR output. 
Of course the actual dictionary lookups could be handled by an external 
program like aspell.

Given that Ocrad is so insanely fast, I think this kind of (optional, 
of course) overhead could be worthwhile. I have been working on a 
larger project with Ocrad for a few days, and while I am pretty content 
with Ocrad's recognition rate, I wish there was an easy way to identify 
and correct the kinds of typical OCR errors which standard spell 
checkers do not know how to handle.

Of course much of this could perhaps be done with a filter on the 
Ocrad-generated text file after the fact, like with a modified aspell 
(http://lists.gnu.org/archive/html/aspell-user/2002-07/msg00003.html). 
But of course doing some of this in Ocrad itself might be beneficial 
because the internal knowledge of the program about the characters 
being worked on could be used. And in terms of programming work, this 
would probably be cheaper than trying to improve the OCR engine 
itself...

Any thoughts on this?

Martin
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (Darwin)

iD8DBQFCyuA4mifxvst1lQIRAgBhAKDJwj2WL6UMkslCSjLDrvNZMQA7GQCdFpYe
6bxhFoCBb7Xw980EIF1t/lM=
=t67Q
-----END PGP SIGNATURE-----