Question

I am trying to implement OCR for my Windows Mobile application. I would like to give ANN a shot but after some searching I cannot find any concrete examples (code) to do this. I have literally no idea how this would work – although I have already read a bit about them.

Can anyone post here some code or specific examples how it can be done ?

Answer 1

Assuming that you have labeled example images of individual characters, and that isolating characters is a solved problem, then the basic steps would be something like:

1. Clean up and pre-process the images (center the character, standardize the contrast, eliminate noise, etc.)
2. Calculate useful features, such as mean and standard deviation of ink pixels, horizontal and vertical projections (row and column totals, respectively)
3. Train a classifier (neural network, linear discriminant, decision tree, etc.) using the features
4. Test on new example images
5. Deploy