Please take a moment to complete this survey below
Library's collection Library's IT development CancelIn this study, comparison of a system for introducing the signature with
kohonen neural network method and feedforward backpropagation networks was
done to know the effect of neighborhood size, alpha and desired error to the winner
index output in kohonen networks. The study also was done to know the effect of
learning rate, desired error and number of hidden layers in the feedforward
backpropagation networks with binary input system that would be changed to the
PCX grayscale 256 level format with 30 x 30 pixel resolution.
The optimum parameters for kohonen networks obtained from this study are
neighborhood size 12, alpha 0.09 and desired error 0.000001 with 50.91% chance to
be succeed. For feedforward backpropagation networks, hidden layer 1 with 18
nodes, learning rate 0.015, and desired error 0.01 gave 43.64% chance to be succeed.
Testing with tampered signature 45% succeeded for kohonen networks and
55% for feedforward backpropagation networks.