Using Deep Learning Networks For Face Spoofing Detection Against Two Types Of Attacks

Face recognition and authentication are already widely used in various applications.Nevertheless,this security system is not entirely safe because it can be fooled by the presentation attack or face spoofing.Face spoofing is categorized into three types,namely photo attack,video attack,and 3D facial mask attack.Until now,almost all face spoofing detection uses handcrafted features or large deep learning networks that are not suitable for mobile phone applications.We use LBP facial images and RGB facial images as input to deep feature extractor.RGB facial images as deep feature extractor input produced high accuracy.However,when the deep feature extractor received LBP facial images as its input,the accuracy is relatively low and unstable.Overall,MobileNetV2 is a reliable deep network architecture in face spoofing detection for mobile phone applications.Main research work and the contribution of this research are as follows:· First,it compares LBP facial images and RGB facial images as input to deep feature extractor.This is done to find out what type of image input has specific characteristics that can distinguish between real access and spoofing attacks· Second,it applies MobileNetV2 to face spoofing detection that never been used before.The accuracy of spoofing detection using MobileNetV2 is very competitive against the use of large-sized deep learning models of VGGNet.· Third,a face presentation attacks database named SSIJRI Face Spoofing Database has been created for evaluating face spoofing detection methods.