Dynamic signature verification system using stroked based features

This paper presents a novel feature-based dynamic signature verification system. Data is acquired from a Patriot digital pad, using the Windows Pen API. The signatures are analyzed dynamically by considering their spatial and time domain characteristics. A stroke-based feature extraction method is studied, in which strokes are separated by the zero pressure points. Between each pair of signatures, the correlation comparisons are conducted for strokes. A significant stroke is discriminated by the maximum correlation with respect to the reference signatures. The correlation value and stroke length for the significant strokes are extracted as features for identifying genuine signatures against forgeries. The membership function and classifier are modeled based on the probabilistic distribution of selected features. Experimental results were obtained for signatures from 20 volunteers. The current 6-feature based signature verification system was calculated to have a false accept rate of 1.67% and false reject rate of 6.67%.