Face anti-spoofing method based on quaternionic local ranking binary pattern features

Currently, face recognition is widely used in many applications including access and security systems. However, a malicious user can spoof such systems by using a photo or recorded video of a legitimate user to gain access. Therefore, it is essential to solve this problem using anti-spoofing measures that can determine whether the face image presented to the system is genuine or a spoof. In this paper, we proposed a novel anti-spoofing system based on local texture features. The local features of the input image are extracted using Quaternionic Local Ranking Binary Pattern (QLRBP). We evaluate the effectiveness of these features using two different classifiers, namely the K-Nearest Neighbour (KNN) and the Support Vector Machine (SVM). Our experiments using publicly available datasets show that our proposed system could achieve up to 95.20% accuracy (at 205.62 ms per frame) using the KNN classifier. Using the SVM, our proposed system can achieve up to 97.36% accuracy at 229.39 ms per frame. These results show that the QLRBP features are suitable to be used in a face anti-spoofing applications.