Dear Shin, I'm happy you have joined us.  Perhaps you can help us
find a solution for this piano roll problem:

When we finally are able to make a high-resolution optical scanner
for music rolls, I also want to preserve the words which are printed
on the margin of the player piano roll, perhaps in a .TIF file or
similar.

But the letters (the "character set") on many piano rolls are
printed with ink brushed through the old-fashioned cardboard
stencil, and the character-recognition software I tried simply
wouldn't recognise the letters!  I made a photocopy of a bit of a
piano roll, and took it to the experts in the Publications shop
where I work at Lockheed.  They were very proud of the Kurzweil text
scanner computer system, which cost over $10,000 at the time.

We tried everything, but it refused to learn: for example, it
insisted that the capital letter "B" must be "13", because of the
narrow spaces in the stencil between the two elements.

Since that time I'm sure there has been much development in software
to perform "trainable character recognition", and I can imagine that
the development was driven by need to scan the three Japanese
character sets.  (Kanji, Katakana & Hirigana -- how's my memory!!)

Can you tell us what's available now that will correctly recognize
plain old English printed with the "stencil font"?

Best regards,

Robbie Rhodes