Abstract:
Biometrics provide a reliable statistical analysis of person's physical and behavioral characteristics for a person's identification and verification. It improves the accuracy of person's identification than other conventional methods such as PIN, password, token etc. Among all biometric systems such as iris, retina, voice, finger prints etc., online signatures provide easy to use, reliable and low cost person's verification system. Most of the previous work done on online signature verification involved statistical Local and Global features extracted from x and y coordinates of the pen. Combinations of different categories of features are less explored in Online signature verification systems. Moreover, Pen's up and down positions along the time series has not used for verification. This motivates us to design a computationally efficient online signature verification system by analyzing online signature characteristics as a biometric trait. Within this context, this work presents efficient Hybrid features and Segmented Local features based methods. Hybrid features based method used DWT based statistics, Fourier descriptors and Global features while Local features based method is applied by using x, y and pen's position (p) values. Euclidean distance and Dynamic Time Warping (DTW) are used for matching/ classification purpose, they first find a threshold value which is user dependent then used for matching. Verification phase consists of two steps; enrollment and matching. In enrollment, reference samples of each user are enrolled while in matching step, questioned sample is matched with stored template. Both proposed methods are evaluated on two datasets; Japanese Online test set and Dutch Online test set. System's performance is evaluated by computing equal error rate (EER), false acceptance rate (FAR), false rejection rate (FRR), accuracy and time in seconds. We achieved 83.112% accuracy on Japanese online test set and 97.45% accuracy on Dutch online test set.
