A Robust and Real-Time Face Anti-spoofing Method Based on Texture Feature Analysis

Face spoofing attack is an attempt to obtain unauthorized access by using photos, videos or 3D maps of an user’s face. In this work, we propose a software-based anti-spoofing method that extracts multiple texture features based on Local Binary Patterns (LBP) in the grayscale and YCbCr color spaces to train binary Support Vector Machine (SVM) classifier, which is then used to classify faces. The proposed method is compared with state-of-the-art methods using Attack Presentation Classification Error Rate (APCER), Normal Presentation Classification Error Rate (NPCER), Average Classification Error Rate (ACER), True Positive Rate (TPR), True Negative Rate (TNR), False Positive Rate (FPR), and Accuracy. Our method performs better than the other state-of-the-art methods when classifying spoofed and non-spoofed faces of the NUAA dataset. In particular, our method presents the smallest FPR, and thus guarantees robustness against spoofing attacks. Furthermore, our anti-spoofing method can be used in real-time applications with an average of 26 frames per second, providing high accuracy with little overhead to authentication systems.