Re: Recognition of Piano Roll Lyrics
By Robbie Rhodes
> Robbie,
> I finally received a copy of Japanese OCR software upgrades and a
> complete manual. With the help of those, I learned the process of
> registering a pattern being corresponded to a charactor. I took an
> example of 'BLUE' with B like as l3. I registered each B, L, U and E
> independently and assigned B, L, U and E respectively. Then I ran an
> automatic recognition process and it recognized those characters
> correctly 100% of time. It appears the registered charactor has
> priority to be recognized first, thus, 1or 3 never resulted. By
> expanding this, I mean to register all 26 charactors, we should be able
> to read the lyrics reasonably well.
>
> Regards,Shin Ohkura
That's great news, Shin, and thank you very much for your efforts. Wayne Stahnke has already attempted to scan the lyric from the piano roll, but the contrast is still too low, and the background noise of the old paper is a big problem. We must still experiment to increase the "signal to noise ratio."
I found an old stencil-punching machine at work, and I made a stencil of all 26 letters plus punctuation marks. The character height is about 10 mm in the original stencil. Then I made a photographic contact print which is very "crisp" with sharp edges, etc., and of course there is no background noise, as in the real piano roll. So this will be a good sample for training the OCR program. I will send these character images to you via fax.
Best regards, Robbie Rhodes
----------------------------------
| Robbie Rhodes |
| Return-Path: rrhodes@foxtail.com |
---------------------------------- |
(Message sent Tue 6 Aug 1996, 14:52:07 GMT, from time zone GMT-0700.) |
|
|
Mechanical
Music Digest™ Archives
