Face Presentation Attack Detection using Color Spaces Features and Convolutional Neural Network

Authentication systems using facial biometrics are currently straightforward to implement. The easy implementation of the system makes it easier for people to carry out spoofing attacks. This spoofing attack can use photos, videos, and other people's face masks. To be able to recognize the existence of this spoofing attack requires good accuracy. One of the efforts to improve accuracy is to improve feature extraction results. Feature extraction using color spaces has been extensively researched. The use of this color space is known to increase performance. Several studies have used several types of color spaces to perform texture analysis. This study will analyze the use of color spaces such as YCbCr and HSV. These color spaces will be combined to get facial features. The color features were input for classification methods such as ResNet50, VGG16, and MobileNetV2 so that they could detect spoofing attacks. Our experiments show this approach's promising result by achieving the lowest EER of 3.62 % in the CASIA dataset.