I would imagine it could be a "Hard Problem". I mean look at all of the problem anyone has ever had with Unicode now imagine trying to reverse engineer it form

I would imagine it could be a "Hard Problem". I mean look at all of the problem anyone has ever had with Unicode now imagine trying to reverse engineer it form images of characters in random font without the bit-stream of the charecter, the benefit of textual context (ie it that a Latin script letter O an zero or O shaped letter from another character set used in a quote name or excerpt from another language), or metadata in file header. I think this will probably be one of those thing that will get better asymptotically until we have sufficiently advanced general intelligence, as alot of it will be context dependent.