[sahana-maindev] Sahana OCR - Image Segmentation Complete!

[Omega S. <ome...@gm...>]

Hi everyone!

I have finished the first phase in segmenting the Sahana OCR form and I have
started work on character identification.

Sahana OCR form should be a web generated form and the final format is not
yet finalized. Therefore, I'm using the draft OCR form (available at
http://www.cse.mrt.ac.lk/~omega/draft.pdf) I developed as the input. I can
distinguish required fields needed for character and data filed recognition
in the draft form. A sample output is available at
http://www.cse.mrt.ac.lk/~omega/sahana_ocr/output.jpg

The source code of my developments is available at
http://www.cse.mrt.ac.lk/~omega/sahana_ocr/sahana_ocr_source.tar.gz. To
compile and run the source, you need OpenCV installed in your PC, available
at http://sourceforge.net/projects/opencvlibrary (For more info,
http://opencvlibrary.sourceforge.net). How to compile and run the source is
specified in a separate file, also included with the source.

The core concept behind the segmentation is to recognize squares/rectangles
in any given image. (Almost all the OCR forms being used today, provide
square shaped areas for the user to enter data, because it minimizes the
errors) Therefore, despite the fact that I'm using the draft OCR form at the
moment, the same code can be used in segmenting the final format of the
Sahana OCR form with very little modification, since it should and will be
based on the square shaped input areas.

The code I have written just uses thresholding, contour tracking and Hough
transform together with some mathematical testing to identify the
square/rectangle shaped contours. The code can be improved to do more
thorough processing to identify squares if needed.

One important thing I realized in doing the development is that we should
have some kind of standard in developing the OCR form. Since the each Sahana
web interface should have the corresponding OCR form, a standard is very
much needed to improve the accuracy. Moreover, a standard will allow users
other than Sahana to come up with their own OCR form adhering to the
standard and use the OCR module to process their forms as well. I'm doing
some homework on this and will start a discussion on this topic in the very
near future.

I would really appreciate any comments or suggestions on my developments so
that I can take corrective actions and improve the functionality of my
work

-- 
Regards,
Omega Silva