On-line signature verification based on forward and backward variances of signature

We propose new features called the forward and backward variances of signature for on-line signature verification. In the proposed method, stable features of signature can be characterized by the forward and backward variances of signature. It is shown that signature can be verified by evaluating the difference of K-L coefficients of the forward and backward variances between the reference and signature to be verified. Experiments are performed on public signature database MCYT-100, consists of 5000 signatures from 100 people. The Equal Error Rate is 4.49%.