Comment by ggm

Because its rendering to bookerly or an analogue a perceptual hash looks like an amazingly good fit. But in general, how applicable would that be to OCR because if you can declare 90% of the text is courier, then it feels like an enormously good way to get over the hump.

I wondered if he was just tuning to the best algorithm for his corner case, but it's one of the algorithms in a decent OCR package anyway?

You'd only have to do a few hint/confirmations.