Among the total number of 330 handwritten signatures,
132 samples were used to train the system and the
remaining 198 samples to test its performance. The samples
were divided randomly into the learning set and the
testing set. All forged samples were included in the testing
set. Fig. 11 shows the correct veri"cation rate obtained
through the experiments.
The experiment (learning and testing) results reveal the
following interesting facts about the system: