I think I'd read the text file of the banner and create a bitmap with a white pixel for every space, then use a black pixel for any other character. Feed it to an OCR module and profit!

Alternatively, I'd read chunks separated by blank lines. I'd then use a hash with the key some form of the chunk (the chunk itself, a digest of it, ...). If it's a chunk you've never seen before, prompt the user for the appropriate character and emit that character. If you have seen it before, just emit the character. For banner text, I think it's better than the previous suggestion, and would make short work of it.

...roboticus