Subject: RE: OCR
From: simo <>
Date: Mon, 12 Jun 2006 16:06:29 -0400

This made me remember that I saw some work on patter recognition that
uses neural networks and Bayesian inference with very promising results,
seem OCR would probably benefit from such approach.


On Mon, 2006-06-12 at 11:03 -0600, Anderson, Kelly wrote:
> I can see where that form of mathematics would be very useful in the
> problem domain. In the "olden days" of OCR, the same was achieved with
> Octal Trees. Probably not as elegant, but the ideas have been around
> forever.
> Progress in Pattern Recognition is very slow. For example, when I took
> my graduate level pattern recognition class in 1987, we used a book
> published in Russia in 1958. It's still one of the better text books out
> there.
> My teacher was rather fond of showing how everything really is the same
> thing. Just to show us one day, he did a mathematical proof that Baysian
> Classification was mathematically equivalent to Neural Network
> classification. I can't remember any of the math, and I'm not sure I
> understood it entirely then, but it was fascinating and intuitive to see
> that it was all really the same stuff...
> -Kelly
> -----Original Message-----
> Anderson, Kelly wrote:
> > Wavelets can be used in Segmentation and Feature Extraction. They may 
> > be very good for segmentation,
> There's been work lately on hierarchical Feature Extraction, where an
> image is analysed, effectively at different resolutions, to extract
> features and their relationships in a scale independent way.  That's not
> for OCR specifically, but for computer vision more generally.
> Cleverly the results are scale independent even though the set of
> resolutions analysed is discrete.  The transforms looked suspiciously
> similar to wavelets to me.
> -- Jamie
> E-Mail messages may contain viruses, worms, or other malicious code. By reading the
message and opening any attachments, the recipient accepts full responsibility for taking
protective action against such code. Sender is not liable for any loss or damage arising
from this message.
> The information in this e-mail is confidential and may be legally privileged. It is
intended solely for the addressee(s). Access to this e-mail by anyone else is unauthorized.