On-line signature verification system

Abstract

A simple but effective automated on-line signature verification is presented. In a study on signatures, it is first established that although the signing process is sometimes variable, the singularities in a signature are stable. The system uses a digitizer to capture as features sampled x and y position signals. A test signature is compared to the claimed reference set and a similarity measure based on statistical correlation is determined using the regional correlation algorithm. The system also accounts for the intraclass variations between signers by using a personalised threshold. A database of 669 signatures is used. The database includes skilled forgeries and genuine specimens of 16 different signers. In a performance test, a correct verification rate of 91.8% was recorded. The system rejected 95.01% of the forgeries and accepted 91.01 % of the genuine signatures. Due to the lack of public databases, these results cannot be directly compared to other existent systems.