An on-line signature verification based on timely nonlinear sampling and sparse representation

A novel approach which generates so-called "potential genuine signatures" by timely nonlinear sampling from an original on-line signature is presented. Velocity vectors of these potential genuine signatures are exploited to construct an user-dependent overcomplete dictionary. Finally, sparse coefficients which served as features are used for verification. There are two main advantages: 1) The DTW becomes unnecessary, which makes the proposed method computationally inexpensive; 2) Multiply potential genuine signatures can be generated from a genuine one, so the difficulties causing by insufficient train samples in a real system are effectively alleviated. Experiments show encouraging results.