A face anti-spoofing approach based on generic sequential model using scale invariant features

Prevailing face biometric-based recognition systems are susceptible to widely attempted spoof or presentation attacks. To countermeasure these attacks, automated spoof detection mechanisms are coalesced with facial access control systems. The commonly adopted approach is to learn a classifier based on handcrafted image features as well recently emerged deep learning models. However, among all, robustness of manually-extracted features and large intra-class variation in facial images are the major bottleneck to the performance. In this paper, we present a face anti-spoofing approach that mitigates the abovementioned issues. Our technique is based on learning a basic sequential model that learns Scale Invariant Feature Transform (SIFT) image descriptor. The technique is based on extracting SIFT features from facial images for training the sequential model that comprise of alternative sets of multiple layers (i.e. Dense, Gaussian and Dropout). The approach is experimentally evaluated on standard and benchmark CASIA-FASD anti-spoofing dataset. Our approach demonstrates state-of-the-art performance in terms of detection accuracy with 90.01 % and a lower HTER of 8.5 %.