On Sun, Apr 19, 2009 at 7:39 PM, Debayan Banerjee <debaya...@gmail.com> wrote:
> Does anyone know of any libraries that can transliterate bengali to
> english. There are tools to the reverse. I need this to solve the last
> remaining road-block in OCR.
> The thing is Tesseract-OCR uses a data structure called
> directed-acyclic-word-graph to store dictionaries for lookup. After an
> OCR has been performed the OCR system matches the output with entries
> in this d.a.w.g. file. Unfortunately the data structure is not suited
> to complex scripts like ours
> <http://groups.google.com/group/tesseract-ocr/browse_thread/thread/5495c4e348a4b272/a6dcfe5d92babb35?lnk=gst&q=dawg%2Bwieghts#a6dcfe5d92babb35>.
> There are 2 solutions. 1) I figure out a suitable data structure that
> handles Indic script and implement. 2) I transliterate the entire
> dictionary and the OCR output to english (26 characters instead of the
> 500 odd for bengali) and then match. I think this should work.


I believe there is a ISO standard for doing this. Take a look at ISO 15919:2001

Saner explanation is at http://homepage.ntlworld.com/stone-catend/trind.htm :-)

There is a thing which does this for Devanagari:
https://www.dealloc.org/~mublin/iso15919.py.html

However, this is not restricted to the 26 characters, but definitely
less than 500 :-)

-sdg-





--
Sayamindu Dasgupta
[http://sayamindu.randomink.org/ramblings]

------------------------------------------------------------------------------
Stay on top of everything new and different, both inside and 
around Java (TM) technology - register by April 22, and save
$200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco.
300 plus technical and hands-on sessions. Register today. 
Use priority code J9JMT32. http://p.sf.net/sfu/p
_______________________________________________
Bengalinux-core mailing list
Bengalinux-core@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bengalinux-core

Reply via email to