[Project-ideas] GSOC OCR

[Sankarshan Mukhopadhyay]

On Tue, Mar 27, 2012 at 3:20 AM, Soumajit Pramanik
<soumajit.pramanik at gmail.com> wrote:

>                       Basically I think I have to first learn well the
> techniques for OCR used by the open source tools. I have to understand why
> this tools are unable to achieve very high accuracy. If I can do that, the
> very next thing will be to decide what extra features I have to add or how
> to scan the letters more accurately. Then I have to use a good machine
> learning algorithm(neural n/w or svm or something more appropriate). Though
> achieving high accuracy like 98% will be very challenging, I do not think it
> is unachievable.
>                        Now, what should I do next??

You should work on a formal proposal that you can submit to the
Melange system. Take a look at the GSoC homepage. Runa had sent an
email to this list earlier about this topic.


-- 
sankarshan mukhopadhyay
<http://sankarshan.randomink.org/blog/>