I've been thinking about this for some time. Finally it dawned on
me.
LilyPond processes text statements & produces the music notation in a
PDF file.
Why not reverse the process and go from PDF (e.g., scanned in), back to
text?
(It's probably a nightmare programming task.) It might require the
scanned image to have alignment marks, and/or be cleaned up in some way.
I think I'd start by developing a program to convert a single line (one
staff), like the attached Bouree, which I created with Lily Pond. I'd
begin by focusing on the clef sign and the key signature. The user might
have to indicate what the first note is, or confirm that the program had
correctly identified the first note.
Just a thought.
Mike Wogan