Re: An improved method for scanning documents

["BLuRry" <brendan.robert@gmail.com>]

I think a lot of the OCR confusion is because there are no well-defined
heuristics that define what the scanned input *should* look like.  If I
remember correctly, most OCR software is trained (in older days, by you
-- in recent days, by the author) on regular english words as test
input.  So, the learning algorithms (neural net, statistical analyzer,
etc) pick up not only what each letter looks like but also in more
advanced systems what letters are likely to come after other letters.
The problem is that such a predictive model also works against you when
working with data that does not resemble the test set.

OmniPage had a training option in early versions, does it still offer
that option today?