check out the website on thinkcycle for other related resources and projects.
-- Gami --- Original Message --- Date: /0/0 From: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> Subject: ThinkCycle>> New Resource: Neat paper on hindi ocr New Resource posted on a ThinkSpace by Sandeep Jain <[EMAIL PROTECTED]> on 2004-01-16. Resource: Neat paper on hindi ocr Topic: Tools for Education ThinkSpace: Reading and Literacy Tool Description: This is a neat paper on Hindi OCR. The description of how to separate a hindi word into characters is the clearest I've seen so far, and that's the central problem in Hindi OCR. I've written to one of the authors to find out if he can make the source code available. I've posted the paper here. If this is a copyright violation, we can remove the paper. The URL to the paper is: http://www.cs.umd.edu/Library/TRs/CS-TR-4519/CS-TR-4519.pdf It may be possible to improve upon the character recognition, based on a technique called Hausdorff Image Comparison, by using a neural network with suitable features. A possible idea for doing is, is the concept of "band density" . If you look at a character as a grid of pixels, and sum up the pixels in horizontal and vertical bands, of different widths (2 pixels, 4 pixels, etc., assuming for example that the character falls on a 16x16 pixel grid), and use each of these sums as an input to the neural network, then it seems that this would be a good classifier. It needs to be experimented with. Sandeep http://www.cs.umd.edu/Library/TRs/CS-TR-4519/CS-TR-4519.pdf See Resource details on ThinkCycle here: http://www.thinkcycle.org/tc-space/tspace?tspace_id=37543 Please forward this note to others interested in contributing to this topic. ======================================================================= To unsubscribe from future postings on this ThinkSpace: http://www.thinkcycle.org/tc-bboard/forum-unsubscribe?forum_id=37541 To manage all your ThinkCycle topic subscriptions: http://www.thinkcycle.org/tc-bboard/subscriptions _______________________________________________ ilugd mailing list [EMAIL PROTECTED] http://frodo.hserus.net/mailman/listinfo/ilugd
