Study of Temporal Variability in On-Line Signature Verification
Résumé
In the field of on-line signature verification, much attention has been paid on classifiers and features used to elaborate authentication systems able to deal with this modality and its specificities such as forgeries and signers variability. Nevertheless, this variability has not really been studied through time as it is, for example, in the field of on-line handwriting recognition. Indeed, most of the databases used were acquired during one or two sessions but there is no database that could describe the variability of the signers through several months. This work presents preliminary conclusions on such database. It seems that there are no trends at all concerning the evolution of signatures. But this study also reveals that the signer variability seems much higher than the one observed in previous databases. A direct conclusion that could be drawn is that performances announced until know could be overestimated considering a real-life exploitation. This should be nuanced considering the experiment size (small sample size and duration) but what seems obvious is that there is a need for further studies and for new on-line signature databases taking into account the temporal variability of signatures