I have written a Guide on Machine Learning with OpenCV, which shows how to use the Multilayer Perceptron class of OpenCV. You can find it as a PDF:
It comes with a sample program, which shows how to work with the OpenCV machine learning API:
I'll update it to OpenCV 2.4 soon and update this page accordingly.
A classical technique is to crop the image just around the letter, resize it to 4x4 pixels, eventually apply a dct to further slim down the search space.(then you could select first 4-6 values). And with those values, train a classifier - a neural network in your case.
Given the width/height ratio of your input, you may choose some different small resolution. Like 3x4, or 4x5.
I advise you to start with digits, or a limited set of characters. Later, you can expand it to the full alphabet, use more complex descriptors, etc.
P.S. If by "working example" you mean "give me teh codez", this is not the right place to ask. One learns best by writing code, reading carefully and working hard.
Good luck!
Asked: 2012-08-13 07:53:45 -0600
Seen: 15,106 times
Last updated: Aug 13 '12
Please be more specific. What do you want to accomplish? What have you tried, what publications did you read about?
I want to work with neural network for recognition characters. I read different articles on the internet. And they often use opencv and fann. But the using described not enough clear. I want to get working example which make me able to teach network by samples and give me opportunities to recognize patterns.