Sorry to contact you there but I do not see an email in your profile.
I just read your comment on derasterize: https://news.ycombinator.com/item?id=25997236
> Based on your project I inferred that adding more matches seems to add more resolution and yeah, outputs are looking even nicer now. I updated the repo!
This is the right approach, but there's a combinatory explosion when computing the matches: you want to select a subset of shapes that in practice are more often right, to save on the matching computation time.
Also, when you are operating with time constraints (ex: video to text), keeping the FPS rate is often more important that the accuracy, as the human eye does a lot of interpolation.
Having different subsets of different sizes and a command-line toggle between them seems like the simpler approach to support multiple uses.
> I also used shapes which are less sharp rather than "perfect match".
Selecting some "less sharp" glyphs is a very daring and unique idea: in retrospect, it should help the bad cases by adding some "blur".
I really like you idea!
Would you be interested in a collaboration to inspect fonts and select glyphs based on 1) their availability (ascii>ansi>unicode) 2) their usefulness (by defining some subsets)