A robust framework for spoofing detection in faces using deep learning

Face recognition is used in biometric systems to verify and authenticate an individual. However, most face authentication systems are prone to spoofing attacks such as replay attacks, attacks using 3D masks etc. Thus, the importance of face anti-spoofing algorithms is becoming essential in these systems. Recently, deep learning has emerged and achieved excellent results in challenging tasks related to computer vision. The proposed framework relies on the extraction of features from the faces of individuals. The approach relies on dimensionality reduction and feature extraction of input frames using pre-trained weights of convolutional autoencoders, followed by classification using softmax classifier. Experimental analysis on three benchmarks, Idiap Replay Attack, CASIA- FASD and 3DMAD, shows that the proposed framework can attain results comparable to state-of-the-art methods in both cross-database and intra-database testing.