Training problem with mis-interpreted chars.

[Jay2]

Hi!

I want to use GOCR to automatize an old DOS Software. Problem is, the characters are misinterpreted.

I've "trained" GOCR, it figured unrecognised chars and when using with the database back (-m 2) it reads all previously unknown chars good, but some where bad at first.

I've try to circonvent the problem by adding manually the misinterpreted chars to the database (I know what they are, and have a screenshots of them), but they still misinterpret it.

I have a sample of every possible characters of this font.

Am I doing something wrong? Is there a way to completly bypass GOCR internal database and use only the external specified database (so it will compare only with my chars). Doing so could also be usefull to implement a "Klingon" database :)

Jay

[Joerg Schulenburg]

two ways for gocr v0.41:

enhigher the certainty by -a 99 or -a 100
or disable the internal engine by -m 128
as described in the example section.

batlh bIHeghjaj

[Jay2]

Hi!

Thanks for your quick reply.

I guess you meant -m 256 to disable the internal engine.

I trained GOCR with -m 386 over a sample of the complete font alphabet, and use it back with -m 258 and it works (almost) flawlessly.

This is a wonderfull piece of software,
Thanks!