On‐line recognition of handwritten signatures by feature extraction of pen movements

Abstract To establish an automatic method of signature verification, this paper presents an on‐line recognition technique of handwritten signatures by comparing the dynamic features extracted from pen movement with the reference data. In this paper, it is observed that dynamic features such as changes in pen inclination and writing force with time are different among the writers and these features are detected by various sensors attached to the pen. In other words, the pen inclination is estimated from the relationship between the pen inclination and the intensity of illumination at the paper surface of light emitted from an LED utilizing a reflective optical fiber sensor. The writing force in the axial direction of the pen is obtained from the relationship between the strain and applied force by using a force sensor installed at the central part of the pen. In comparison with the reference data, simple features such as number of pen‐ups and pen‐downs and maximum, minimum and average values of time histories of the writing force and pen inclination are considered first. If the data are rather significant, then further comparison is made with respect to the time series waves of the pen inclination and writing force before a final decision as to whether or not a signature is authentic is made. Finally, the proposed technique is applied to a set of test data consisting of autographs and forgeries and it is shown that this method distinguishes an autograph from a forgery from the changes in writing force and pen attitudes even though two signatures resemble each other.