Online Signature Verification Based on Stable Features Extracted Dynamically

Abstract

Effective information extraction and optimal feature subspace selection have great effects on the performance of online signature verification. One of the difficulties about online signature verification is to extract and select effective features, which are stable in genuine but discriminative to distinguish forgeries. Different from other works, stability of spectral information inherent in signatures is analyzed in this paper, and stable spectral information is selected dynamically dependent on individuals to reconstruct the stable spectral features. In order to extract more effective spectral information, features are decomposed by wavelet packet with the optimal mother wavelet. To enhance the security level of online signature verification, discriminative capabilities of stable spectral features are analyzed by factorial experiment design. The optimal feature subspace is selected according to contribution rate. Furthermore, we proposed a simple and effective modified dynamic time warping (DTW) with signature curves constraint to solve the problem of heavy computation of DTW. Several experiments are carried out on open access database of MCYT_DB1 and SVC2004 task2 which consist of 6600 signatures from 140 individuals in total. Experiment results demonstrate the effectiveness and robustness of our proposed method.