Effect of the neural network size on performance of authentication systems for fingerprints

We constructed multi-step authentication systems of layered neural networks for fingerprint, where we can deal with a number of patterns and can control authentication rates by the number of step in the system. We clarified the effect of neural network size on performance of the authentication systems for fingerprints. From the results, we found that the authentication rates for 1 step with neural networks size of 9(number of input layer units)-9(number of hidden layer units)-1(number of output layer units) and 16-16-1 were 99.852% and 99.985%, respectively. The multi-step authentication system consisted of the neural networks which were arranged in series. Therefore, the authentication rates of the system were expressed by the equations of {100-(0.148)^N}% and {100-(0.015)^N}%, where N is the number of steps.