Re: start of SIRAGI project

[Tarik FDIL <tfdil at sagma dot ma>]

Salam,

If I summarize our discussion, here is the main points :

- tif files are not in their right place in SIRAGI, so they should  be removed 
from CVS.

- GOCR has a poor design, so it can not be adapted for supporting arabic. It's 
not a  bad idea to write another OCR from scratch.

- SIRAGI-OCR should not make the same mistake as GOCR : supporting one type of 
script. Its design should be general and good to support Arabic and other 
scripts (latin, etc.)

Thanks for these recommandations. I will update CVS later and give more 
details in todo list. 

Mete wrote:
> [...] Are you planning to base SIRAGI completely on top of
> neural networks? It seems like that would be the way to achieve a really
> efficient and flexible OCR.

There is two important parts in an OCR : line/character detection and 
character classification.  The former part is implemented in SIRAGI using a 
finate automata. It is not an easy problem... The latter one will be coded 
using a neural network. The nn is designed, trained and tested using Stutgart 
Neural Network Simulator. 

I think we should write at least two engines based on neural networks : the 
first engine get pixels as input and the second get curves as input. When the 
two engines agree on a character, the recognition is confirmed.  Currently we 
are working on the first engine.

Kind regards
Tarik

Le Vendredi 8 Avril 2005 08:48, Mohammed Elzubeir a écrit :
> Behdad Esfahbod wrote:
> > That's exactly the point.  Just use libtiff, libpng, and libjpg
> > and you have supported all formats that matter.  Having libtiff
> > files in your source code just makes it look bigger than it
> > really is.
>
> I cannot judge the developer here for why he made certain decisions but
> this seems to make sense to me (use pre-existing libraries and link to
> them instead).
>
>  > [..]
> >
> > So, please design your's with multiscript support in mind.
>
> I think this is the crux of the matter. The mistake we often make (not
> always) is that we end up doing Arabic solutions, _because_ other
> applications are not i18n'ized enough. Yet, we do the same thing the
> original authors did -- stick to their own cultures and not think of the
> rest of the world.
>
> I don't know about SIRAGI itself, but I can say that in the case of
> Duali, the most likely conclusion for it _will_ be integration into
> other mainstream applications.
>
> Behdad, need I remind you that it took OVER A YEAR for VIM's author to
> incorporate the Arabic patches? We can't just sit around waiting for
> mainstream developers to integrate our work and neither do we want to
> start forks unless we feel that the maintainers have no intentions of
> doing anything about it. After all, the chances of the success of a fork
> is mostly not very high.
>
> Enough said, let's get back to work. Tarik ,if you can please expand on
> your TODO list so others who are interested can pitch in work, it would
> be great.
>
> Regards,
> Mohammed Elzubeir

-- 
Tarik FDIL
Service informatique
SAGMA
GSM (212) 61 14 34 49
Tél. Bureau (212) 22 44 07 17,  (212) 22 31 64 69
Fax : (212) 22 44 63 52
Web : professionnel : http://www.sagma.ma
personnel : http://www.sagma.ma/tarik/
paternel : http://www.sagma.ma/oapam-casa/
                http://siragi.sourceforge.net