Seems like a similar but less elegant solution as parsing and normalization to a “safe” subset but not just blasting it to pixels.

https://github.com/caradoc-org/caradoc

http://spw16.langsec.org/slides/guillaume-endignoux-slides.p...