Rendering followed by OCR and making sure that the computer doesn’t see more or less than the user does. Tricky and computionally more expensive.

And even then. I might question if what is rendered and then OCR is same as humans see on their screens... I am pretty sure there will be some tricks to change things enough for computer to get something different from humans.