On Fri, 1 Feb 2019 20:23:26 +0100, Peter Lebbing wrote: > On 01/02/2019 17:37, Stefan Claas wrote: > > Tesseract did not do a good job, to many errors. > > Just an idea: OCR'ing a special OCR font like the two classics I > mentioned will go a lot better if the OCR engine *knows* it is looking > at that font. They designed the glyphs to be dissimilar. I don't know if > there are any free software OCR engines that can restrict themselves to > a specific font, I'm just reasoning about it without domain knowledge. > > Also, if you choose an encoding that avoids similar glyphs like one and > ell, zero and oh, etcetera, your miss rate should go down.
Well, i googled a bit and it seems one has to train tesseract to give good results. As understood Google's engine uses also tesseract, but it must be trained then pretty good, i assume. > > Then i googled a bit and ... Google can do it. > > That doesn't seem useful for secret letters. And I don't think you'll > get an offline engine which has been trained like theirs from them. Probably not, but i thought to share my findings. > PS: Could you removed the (was: ...) bit from the subject in replies? I > think I'll stop doing that type of formatting from now on. I saw it > being used quite some time back and when it works it's okay, so I > followed suit. But it's not working that well anymore. Sorry, i always overlook this ... Regards Stefan _______________________________________________ Gnupg-users mailing list Gnupg-users@gnupg.org http://lists.gnupg.org/mailman/listinfo/gnupg-users
