Neural approaches for human signature verification

This paper describes three neural network (NN) based approaches for on-line human signature verification: Bayes multilayer perceptrons (BMP), time-delay neural networks (TDNN), and input-oriented neural networks (IONN). The backpropagation algorithm was used for the network training. A signature is input as a sequence of instantaneous absolute velocity (|/spl upsi/(t)|) extracted from a pair of spatial coordinate time functions (x(t), y(t)). The BMP provides the lowest misclassification error rate among the three types of networks.