optical scanning, cont. (133)

Fri, 7 Apr 89 20:27:28 EDT

Humanist Mailing List, Vol. 2, No. 812. Friday, 7 Apr 1989.

(1) Date: Thursday, 6 April 1989 2251-EST (38 lines)
Subject: Scanning on KDEM 3

(2) Date: Fri, 7 Apr 89 10:29 SET (53 lines)
From: Lelio Camilleri <CONSERVA@IFIIDG>
Subject: RE: ocr for music

(3) Date: Fri, 7 Apr 89 17:07:39 EDT (17 lines)
From: bobh@phoenix (Robert Hollander)
Subject: Re: optical scanners, cont. (94)

(1) --------------------------------------------------------------------
Date: Thursday, 6 April 1989 2251-EST
Subject: Scanning on KDEM 3

Since Jonathan Altman has made reference to our work at CCAT
(Center for Computer Analysis of Texts) on the KDEM 3 scanner,
let me clarify a thing or two and promise to add more later,
if anyone cares. Once Willard and Ian tell me that I've done my
duty regarding the Toronto Conference Booklets material, and the
Yearbook, I will return to more leisurely discussions of various

We found that the KDEM 3 was probably slightly better than the
KDEM 4000 for our multilingual needs -- certainly no worse, if
perhaps slightly less streamlined in operation. We have not scanned
Arabic or Syriac or similar ligatured scripts, although we have
played around with them. We have had very good success with
Hebrew (even some pointed), Aramaic, Yiddish, Coptic, Greek,
Russian, and modern European typefaces -- depending, of course,
on the quality of the material being scanned. The strategy that
Jonathan alludes to, which we have used to great advantage, is
letting the KDEM guess what it is "seeing" (e.g. Hebrew "B" looks
like a "2" and Hebrew "R" looks like a "7"), forcing it to clarify
any ambiguities, and working from there. A post-scanning tailor
is used to convert the jibberish to what you want.

When there is time, I will quibble with Jonathan's preference for
online correction/editing of what is being scanned. For many large
jobs, our experience is the opposite -- the operator will take much
more time than is practical to correct as the machine scans. So we
use a battery of post-scanning tailors, formcheckers, etc., to
deal with this situation. Uninterrupted, the KDEM will scan between
30 to 60 pages per hour of most printed or typed materials. I would
guess that online editing would take at least 3-5 times as long with
most of the materials we have scanned. (And the operator would need
to know the language quite well.)

Enough for now. Bob Kraft (CCAT)
(2) --------------------------------------------------------------56----
Date: Fri, 7 Apr 89 10:29 SET
From: Lelio Camilleri <CONSERVA@IFIIDG>
Subject: RE: ocr for music

To answer a query on music encoding by ocr, I enclosed an excerpt of
an article of mine about computer applications to musicology in which
I report information on this subject. Useful information can be found
in the 1986, 1987, 1988, and the forthcoming 1989 issues of the
Directory of Computer Assisted Research in Musicology, published
by the Center for Computer Assisted Research in the Humanities,
525 Middlefield Road, Suite 120, Menlo Park CA 94025,
(415) 322-7050.
Lelio Camilleri
Conservatory of Music L. Cherubini
Musicological Division of CNUCE-C.N.R.
Florence, Italy.
As to the research in optical character recognition for music, we
are in the initial stage even some studies dated back to 70s. At
present time, two Japanese teams are working on optical scanning
for music with some successes. The team of Prof. Ohteru at
Watseda University has been developing a strong competence in
automatic recognition of printed music and its translation to and
from Braille. The reading of score is one of the task they are
pursuing in their research in robotics.
Prof. Inokuchi's research group, Osaka University, has worked
out, as a part of a more complex expert system for music, a
recognition system for printed piano music. It is carried out
into two phases, pattern recognition and semantic analysis phase.
The rate of accuracy is 94.2% for Beethoven's "Fuer Elise", and
89.3 for Chopin's Etude No.3. The average processing time is 90
minutes a page.
Another current research in this subject is the one carried out
by Alistair Clarke, B. Brown, and M. Thorne (University of
Cardiff, UK) on an inexpensive optical character recognition for
music notation. Their system currently identifies a stave of a
single melodic line. The symbols the system recognizes are
accidentals, clefs, key, and rests. Further developments concern with
the identification of the other musical symbols as well as
overlapping symbols and spurious characters to be ignored in the
recognition process. The system runs on an IBM PC, demonstrating
that is possible to use OCR techniques on a personal computer.
Nicholas Carter thesis (University of Surrey, Guildford)
completed a thesis on optical scanning of music, one of several
similar efforts in 1989. The system converts the captured musical
information to one of the code used by a commercial program for
printing music. Other works on automatic transcription have been
developed by Andranick Tanguiane of the Soviet Academy of
Science, and Neil Martin (Thames Polytechnic, London).
(3) --------------------------------------------------------------29----
Date: Fri, 7 Apr 89 17:07:39 EDT
From: bobh@phoenix (Robert Hollander)
Subject: Re: optical scanners, cont. (94)

To add a note to Jonathan Altman's useful communication, I should first
of all say that Jonathan has done a *very* great deal to bring the DDP,
of which I am the director, into its present condition. He is a tremendous
resource. I want to emphasize what he said in his last remark: unattended
scanning (unless one happens to hit a perfectly "easy" text--and even then
there are problems) is *not* a good idea. Further, our project spends its
most expensive dollars on editing, which involves people who have to be
trained dantists. Most KDEM (or other scanners) can be operated by people
with less in the way of qualification. Paying them to do more work, thus
creating cleaner text, saves considerable money at the editing end.

Robert Hollander