Yeah, if I were doing it I'd probably tend to be pretty cautious - if I see any popularly-used typeface that would confuse the characters I'd mark them as homographs.

It would absolutely have to be a pretty labor-intensive, manually-maintained database.