Optimal Mother Wavelet Selection for On-line Signature Verification

Abstract

On-line signature verification is one of widely acceptable biometrics. Since it can reveal more intrinsic characteristics inherent in writer, spectral information that contains in features can be used to on-line signature verification. In order to extract more effective spectral information, features are decomposed by Wavelet packet (WP) with optimal mother wavelet. Information entropy theory is introduced to reveal the disorder of spectral information contained in signature. The optimal mother wavelet, which is used in WP analysis, is selection based on maximum energy-entropy values to extract more effective information of signature to be used in verification. Because of different writing habits, stabilities of spectral information inherent in signatures maybe varied. To improve the robustness only stable spectral information is extracted. Several experiments are carried out on standard on-line signature dataset MCYT_Subcorpus_100 (DB1), which consists of 5000 signatures from 100 individuals in total. Experiment results demonstrate that the efficiency of on-line signature verification is improved greatly by our proposed method. The best result is given by EER=3.11%, which also indicates the effectiveness of our proposed methods.