Authors: Joseph Shelton; John Jenkins; Kaushik Roy
Addresses: Department of Computer Science, North Carolina A&T State University, Greensboro, NC 27401, USA ' Department of Computer Science, North Carolina A&T State University, Greensboro, NC 27401, USA ' Department of Computer Science, North Carolina A&T State University, Greensboro, NC 27401, USA
Abstract: Replay attacks on biometric based access control systems are dangerous if counter-measures are not instilled to mitigate these attacks. An attacker can capture packets along an unsecure network and replay them at a later time to gain unauthorised access. Traditional biometric systems use a single feature extraction method to represent an individual, making captured data hard to change. Previous work has been done to represent an individual using a set of randomly chosen, unique extractors for each access attempt. However, there are a limited number of unique extractors that can be created. This work seeks to extend the number of unique representations that can be created by evolving a set of unique masks to apply on unique representations of templates. The results on this work show that templates with masks applied on them are unique enough from one another to mitigate replay attacks while maintaining high recognition accuracy.
Keywords: periocular recognition; genetic and evolutionary computations; feature extraction; feature selection; cyber security; biometrics; artificial intelligence; replay attacks; privacy; authentication.
International Journal of Information Privacy, Security and Integrity, 2017 Vol.3 No.2, pp.96 - 116
Available online: 07 Dec 2017 *Full-text access for editors Access for subscribers Purchase this article Comment on this article