4 ms·

> "I decided to create a typeface that would be unreadable by text scanning software With enough traning, any typeface is OCR-able

• iso8859-1 13y ago
  What if the kerning is non-deterministic? Identifying letters is not enough, you have to know their order too. For example, say my kerning flips every second letter. OCR alone can't solve that problem, you need AI.
• sramsay 13y ago
  I'm sorry, but I've spent about twenty years trying to deal with digital versions of texts written before 1900, and I think that's just bullshit. This problem is still really, really fucking hard, and still really, really unsolved. If there's an existing OCR program that you can just "train" using, say, clean scans nineteenth-century newspapers, and have it have a success rate anywhere above "completely sucks," then I'd like to see it (Tesseract, certainly, is reduced to tears by such things -- in fact, Tesseract is reduced to tears by the typeface used in the Federal Reporter). I've used every OCR program ever written. None of them come remotely close to doing the job. I agree that this is theoretically solvable, but it's a little precious to hear people say over and over how "do-able" something is when no one has actually done it.
  • iso8859-1 13y ago
    Anything humans can do is theoretically solvable, isn't it?