A time varying vector autoregressive model for signature verification

This paper examines a new approach for verification of on-line handwritten signatures. A signature is treated as a vector random process whose components are the x and y Cartesian coordinates and the instantaneous velocity of the recording stylus. This multivariate process is then represented by a time varying p^th order vector autoregressive (VAR) model which approximates the changes in the complex contours typical in signature analysis. The vector structure attempts to model the correlation between the signature sequence variables to allow the extraction of superior distinguishing features. The model´s matrix coefficients are used to generate feature vectors which permit the verification of a writer´s identity. An experimental study is presented which compares performance with a 1-D counterpart