A model for comparing signatures

A statistical model for signature recognition and verification by computer, using data obtained online from a digitizer, instrumented pen, or similar device, is proposed. The model recognizes that repeated test signatures by the same writer are similar but not identical. The model considers a mean or average signature that is independent of speed of writing, location, scale, and orientation, together with a measure of variability of the latter four factors. Various methods of estimating the mean (average) and variance of a writer's signature are proposed.<
>