Page 1 of 1

Optical character recognition

Posted: Thu Oct 21, 2010 2:35 pm
by josh
Image

I want to find the x, y coordinates, or bounding box of where the numbers within the circles are located. They won't always be laid out like this or in circles. Ideally I should get some kind of output like

'1' is located at 200px X 300px lower left and 250px x 450px upper right
'2' is located at 220px X 320px lower left and 270px x 470px upper right
etc...

is there any good open source or commercial tools for processing images of these sorts?

Re: Optical character recognition

Posted: Thu Oct 21, 2010 4:27 pm
by VladSun
Well, in this case it's relatively easy to find these coordinates. You just need to find the originally used font (i.e. Arial, Verdana, etc.) build a set of different sized number and perform correlation analysis.
I bet this will achieve what you need.

Re: Optical character recognition

Posted: Fri Oct 22, 2010 11:29 am
by Jenk
Never used it, simply searched for "opensource ocr" and this came up:

http://code.google.com/p/ocropus/

Re: Optical character recognition

Posted: Fri Oct 22, 2010 5:25 pm
by Weirdan
Jenk wrote:Never used it, simply searched for "opensource ocr" and this came up:

http://code.google.com/p/ocropus/
Wouldn't work out of the box (I tried :) ), and has pretty scarce documentation. Perhaps it could be trained to recognize what josh wants, but it doesn't seem to be an easy task.