On this lazy afternoon, I was thinking about how make an algorithm to read letters from a picture, screen. Some one heard something about it?

It''s called OCR (Optical Character Recognition). www.google.com should give you lots of links.

botman

There are a bunch of ways to do this. The simplest is a template matching system, where you make an image of each letter and scan it across the target image, doing a pixel-wise multiplication and summing the products. Good matches for the letter will return high values while non-matches will return low values. This method is very inefficient and rather vulnerable to noise, scaling, and other things, but it''s very easy to implement and can work pretty well.

A more elegant approach is to divide the image into connected regions (i.e. letters) and find a signature for each region. Then, you can compare each signature to those in a pre-made database of letter-signature mappings to bring out what letter each region is. This is a bit harder, since finding connected regions can be a bit of a pain (not too hard though), and finding a good signature algorithm takes some experimentation. Try searching google for chain codes, or something like "image recognition signature" for some details.

Just some thoughts, but they should get you started.

You can also look up Independent Component Analysis (ICA) and Artificial Neural Network (ANN) methods for OCR.

Good luck,

Timkin