Presentation Attack Detection for Sclera Biometric Applications

Reliability of a biometric system is judged by its ability to recognise genuine samples, reject impostors and the ability to withstand attacks in any form. A recognition sub-system does the job of accepting a genuine user and rejecting a non-user. The attack detection system tries to identify spoofs so that it can ensure that the sample obtained is from a live person. Proposed attack detection strategies can be used in ocular biometric recognition; more specifically, sclera recognition. Sclera recognition needs multi-angle eye images to capture larger portion of visible sclera. The process of obtaining multi-angle eye images is itself an active spoof detection strategy. It is performed by using a supervised eye-gaze detection deep learning model. In addition, passive spoof detection strategy detects fake images by examining the nature of image captured. Such images are classified by using a DenseNet based supervised deep learning model. Individually both these subsystems achieves about 99 % accuracy. Collective implementation of both these systems gives about 100% accuracy. The proposed attack detection strategy can be implemented in a smart phone with ocular biometrics.