Scanning of devanagari

[Peter M. Scharf]

A Devanagari OCR program was developed and distributed by Jeff
Turnbull and Julian Miller, then of Maharishi Vedic University.  The
program achieved a 95% accuracy on printed text when trained on the
text itself if the text did not have much variation in vertical space
(between lines).  However, the time and energy necessary to train the
program, scan the text and correct 5% of errors proved unproductive
so that the developers themselves turned to double-keyed input
instead.  Dominik's point must therefore be reiterated.

>Devanagari OCR still can't be done as cheaply and accurately as
>double-keyed manual entry.  That's for good printed text.  Manuscripts are
>probably never going to be readable automatically.  The material at
>http://www.cedar.buffalo.edu/ILT/ is pure fantasy, in my view.

One should nevertheless be aware of current attempts to improve upon
scanning technology being carried out under the auspices of the
Technology Development for Indian Languages Program of the Govt. of
India: http://tdil.mit.gov.in/

--
**************************************************
Peter M. Scharf             (401) 863-2720 office
Department of Classics      (401) 863-2123 dept
Brown University
PO Box 1856                 (401) 863-7484 fax
Providence, RI 02912        Scharf at brown.edu
http://www.brown.edu/Departments/Classics/Scharf/
http://sanskritlibrary.org/
**************************************************