Incremental Support Vector Machine for self-updating Fingerprint Presentation Attack Detection systems

In this years Fingerprint Presentation Attack Detection (FPAD) had an increasing interest and the performances became acceptable, especially thanks to the LivDet protocols into the International Fingerprint Liveness Detection competition. A security issue arose from LivDet2015: the FPAD systems are not invariant towards the materials for fabricating spoofs. In other words, some previous works pointed out the vulnerability of these systems when an attackers uses unexpected materials. In this paper, we proposed a solution that exploit the self-update abilities of the classifier to adapt itself to never-seen-before attacks over the time. Experimental results on four LivDet data sets showed that the proposed method allowed to manage this vulnerability.