Just finished porting this applet to HTML5. It also works on an iPad. You can see the new example here.
This applet demonstrates simple handwriting recognition using a Kohonen Neural Network. Try it out! Draw any capital English (Roman) letter and click "Recognize". The applet will attempt to tell you what letter you drew. The program will not be 100% accurate, probably closer to 70%. Additionally it was trained on my handwriting. Want to learn more about Kohonen Neural Networks? Read about them in my book.
Want to try out the Network with your handwriting? Or perhaps the 10 digits? You can retrain the applet for any characters you like. First click "Delete All". This clears out the neural network. Now you must add the letters. Draw your first letter, just like you did when you were trying to recognize a letter. Type the letter you just drew next to the add button and click the add button. Your letter will now be added to the Known Letter Database.
Continue this process until you wish to add no additional letters. Now click "Train". The Neural Network will train almost instantly. It is now ready to recognize your characters. Your changes are local only. No information is transmitted back to Heaton Research, so your training will not be saved.
This example uses a small neural network designed to function as an applet. If you would like a more "industrial strength" neural network framework consider the Encog Machine Learning Framework. Encog is available for both Java and .Net, and includes an OCR example just like this one.