Stumbled upon a toy which performs Japanese handwriting recognition online by using ajax as the frontend. As I said it’s just a toy so the correctness is so so, but nonetheless a very good proof-of-concept. This toy uses Support Vector Machine as the underlying algorithm so the program is able to learn from the user’s handwriting. Ever since I made lamerpad I realize that pattern recognition should be a better means to handle handwriting recognition than analyzing stroke patterns, such as stroke order, numbers and positions, which lamerpad employs, as stroke structures vary from person to person. But by using machine learning algorithms like SVM or artificial neural networks, a hybrid system could be made, such system not only can learn from how the character looks, but also how the character is written by the user. By now Chinese handwriting recognition technology is already quite mature, but just no such open source programs, SCIM handwriting module, anyone?
- http://www.teresi.us/html/writing/handwriting.html: “Handwriting Recognition, Using a Neural Network Character Classifier”, have a description of Newton Messagepad’s handwriting recognition engine
- http://www.byte.com/art/9609/sec17/art7.htm: overview of companies that work on Chinese handwriting recognition