Stroke encoded Chinese handwriting input system based on back-propagation networks

We have developed an on-line Chinese handwriting input system using artificial neural network technique. A digitizer tablet with a stylus is used for writing Chinese characters as input to the system. The character strokes and their writing sequences are captured by the system. A backpropagation network is designed and implemented to recognize and classify these basic strokes rather than the whole set of Chinese characters. Due to the simplicity of the basic strokes, a backpropagation network with few neurons is sufficient for training and recognizing purpose. A new stroke encoding rule is proposed for representing a Chinese character in terms of some basic strokes. This system attained a recognition rate of 94% for basic strokes and an overall recognition rate of 70% for Chinese characters within 1 second using a personal computer.<
>