Skilled Forgery Detection in On-Line Signatures: A Multimodal Approach

Signature recognition has a long history of usage in authentication of transactions and legal contracts and hence is easily accepted by users in a variety of applications. However, the problem of skilled forgeries is an important challenge that needs to be overcome before signature recognition systems will become viable in unsupervised authentication systems. In this paper, we present a multimodal approach to forgery detection, where a physiological trait, the face of the signing person, is used to validate the signature. Methods of normalizing and combining the matching scores from the individual modalities are investigated. Test results of the system on a database of 100 users is presented. The system achieves an equal error rate of 2.2% in the presence of high quality skilled forgeries and could detect all the skilled forgeries at a genuine acceptance rate of 75%.