Archived

This topic is now archived and is closed to further replies.

VisualB4BigD

Natural Handwritting After-Processing Recognition

Recommended Posts

VisualB4BigD    122
I am making presentation software and I want to add the ability of Handwritting recognition. But I don''t want to have to create or use a "Grafitti" type recognition algo. I want to be able to take a written word like, "Dog" and have the computer determine that it says Dog. Of course it would take alot of training to get useful and reliable results. We could write the letter "t" on the screen and I should be able to write it as a normal looking t and not something that has to be one connected line. That''s why I need the computer to process the character after it''s been written, which means I won''t have any points or lines to use for processing. All I will have is where the character is on the screen and how big it is. Can anyone point me in the right direction. I really want to make this happen. -Thanx Guys (Gals)

Share this post


Link to post
Share on other sites
eldee    122
ok, im just gonna ramble some theory here, since i''ve never
attempted anything like this before...

you could have a series of training episodes where the user
was asked to write in some words for calibration.
using that data, you could create a feathered mask of what
the letter would look like (when i say feathered, i mean the
mask wouldnt be EXACTLY like the letter, but it''d leave some
room for error.)

when you''re actually using the software, you could have some sort
of bounding box that was over the letter after it was drawn,
then you could compare it to all possible characters.. closest one is a winner.. just some ideas, im really just spewing here.



-eldee
;another space monkey;
[ Forced Evolution Studios ]


::evolve::

''In C we had to code our own bugs. In C++ we can inherit them.''

Share this post


Link to post
Share on other sites
eldee    122
ok, here's a diagram ()
imagine the black spots as your masks and the red text as
what the user would write in.


[edit: i just re-read part of your post and is says all you'll
have is where it is and how big it is!? im afraid you'll need
more information to detect handwriting]

PS: im pretty sure this is the method WindowsCE: Transcriber works.

-eldee
;another space monkey;
[ Forced Evolution Studios ]


::evolve::

'In C we had to code our own bugs. In C++ we can inherit them.'

[edited by - eldee on October 3, 2002 4:15:24 AM]

Share this post


Link to post
Share on other sites
Timkin    864
The problem you are trying to solve is called ''Optical Character Recognition'' (OCR). You can find tonnes of information regardings techniques for OCR on the web. One of the quickest and easiest techniques for OCR involves Independent Component Analysis (ICA), which treats the pixel map as a matrix and determines a minimal set of eigenfunctions that characterises the matrix. This characterisation forms the basis of identifying which class (character) a map belongs to. During the 80''s a lot of work was done on training ANNs for the task of OCR. There''s tonnes of literature on the web regarding OCR techniques, so it should keep you busy for a while!

Good luck,

Timkin

Share this post


Link to post
Share on other sites