Facial spoofing indicator using deep learning

Abstract

Moving along with advancements in the technology sector, biometric verification is becoming more and more common due to its simplicity and user-friendliness. Out of all the biometric verification, facial biometric verification is the most common. Facial biometric is linked with an increase in vulnerability to facial spoofing attacks as it is easy to acquire individuals’ photos from platforms such as social media or Google. Therefore, the aim of this project is to find ways on how to improve the current deep learning models by approaching photo attacks. Photo attack datasets were used to train the model and to test its accuracy by classifying 2 classes of images into fake and real. With the usage of RandAugment [20], it shows that the models can produce slightly better results than normal data augmentation.