Hybrid Deep Learning Approach For Face Spoofing Detection

Abstract

For access control in different applications and in their authentication, Face Recognition plays a vital role replacing the traditional password methods. To accurately simulate the details of physical and physiological users, crime experts are developing techniques; these are known as spoofing attacks. Along with the traditional biometric techniques robust counter measures should be introduced to prevent such thefts. For the work, deep features are extracted from the image with the help of the modified CNN (Wavelet CNN). A stacked auto encoder is introduced for spatiality reduction. Along with the hybridisation, type of spoof attack is also detected in terms of printed attack or camera attack. In order to increase the detection rate, a score based prediction is also performed presenting accurate results in finding the type of attack.