Comment by audiometry

I'm surprised that handwriting recognition is such a problem for you. Wonder if you can improve accuracy by customizing the training for each user as you grow a larger corpus of examples in each users' hand?