Practical issues in enrolment validation for robust automatic signature verification

Despite the investigation of many different possible candidates for adoption in the verification of an individual's identity, the handwritten signature is still the most widely used and generally socially acceptable biometric for this purpose. It is the issue of developing an appropriate signature model which forms the principal focus of the work presented, since this is clearly critical to the success of the overall system performance. If the model adopted encapsulates features of the sample set which result in an atypical representation of the signer's execution pattern and/or an atypical distribution of observable visual features of the signature image itself, then subsequent performance in operation may be severely compromised. This can result in a substantially less secure or user friendly system, which may on the one hand accept a greater number of invalid signature samples as genuine or, on the other hand, may result in an unacceptably high false rejection rate for genuine signatures. In either case, system performance will then fall below that which might be required in an optimal configuration. In the work reported, exactly this problem-of optimal reference model construction-is addressed through a process of signature model "validation" during the enrolment period. Other approaches have been implemented (W. Nelson et al., 1994) to attempt to validate candidate enrolment signatures as they are donated, but this has generally been realised as a manual process and at the discretion of the contributing signer.