Writer verification using multiple neural networks

A writer identification system is described an this paper based on the use of multiple neural networks. In this system a set of Chinese characters are written by the people who are registered in the computer. One neural network is trained for all samples of each character. In the testing mode, a person is asked to write the same set of characters and the decisions of all neural network are combined to identify the person. The method has been verified to work highly reliably on real testing data and it compares favorably with the distance based method proposed by other researchers.