SVM Based Online Signature Verification Technique Using Reference Feature Vector

Abstract

A new algorithm for online signature verification has been presented in this study. The eighty features are identified and used to characterize the online signatures. The signatures are further represented in a vector form by using symbolic representation. In proposed algorithm, two Reference feature vector for both genuine and forgery signature vector are generated for training purpose by the use of random intervals. The two class SVM classifier has been implemented for the verification task. The rigorous experiments by generating nearly a total of 1,000,000 training and testing files are performed to check the performance of the system. The popular benchmark database from first signature verification competition held in 2004 is considered for experimentation. The results are shown in terms of FAR and FRR. We have obtained a minimum value of mean FAR and mean FRR as 7.0092 and 2.2633 respectively for the same dataset in comparison to previous work. After performing sixty more iterations, we have evaluated sixty such values of FAR and FRR by changing the width of the original interval up to sixty times and analyzed behavior shows that if we keep on increasing the width of the original interval, the value of mean FAR and the corresponding FRR will keep on decreasing to the minimum level.