OCR'ing the page instead of reading the 1 pixel font source would add another layer of mitigation. It should not be possible to send the machine a different set of instructions than a person would see.