On Analysis of Face Liveness Detection Mechanisms via Deep Learning Models

Abstract

In recent times, susceptibility of face recognition system to spoofing attacks has received a significant attention from research community. These attacks simply involve presenting an artifact (i.e., video replay, print photo or fabricated mask)to the sensor component and have shown to be capable of deceiving face recognition (FR) systems. The design of an anti-deception method that is termed as face spoof detector is a challenging task that aims to reveal a fake user seeking to mislead the verification system. In this study, we present an analysis of state-of-the-art face spoofing attack discernment techniques along with a taxonomy. A focused survey of face anti-spoofing via deep learning-based methods with special emphasis on latest trends in deep learning techniques is expounded. Additionally, a comparative summary of benchmark face-anti-spoofing datasets employed for various data-driven models is also illustrated. We offer a brief overview of various evaluation protocols for measuring the effectiveness of FASDD approaches. The presented study investigates several key challenges that are open to researchers for further progression in this active field of FLD. Our analysis clearly advocates that among all, accuracy of FLD algorithms in cross-material scenario is still a challenging task. The training overhead of deep convolutional neural networks (CNN) deployed as anti-spoofing detectors demonstrates comparatively better accuracy with an additional training overhead.