Face Anti-Spoofing in Face Recognition: Using Color Texture and Corner Features with Mtcnn and Facenet on Raspberry PI
Almer A. Lamsis, John Daniel M. Osmeña, Noel B. Linsangan · 2024
Face recognition, a dependable biometric technology, utilizes facial information to verify an individual's identity. In the field of research, the emphasis of face recognition has been on enhancing the recognition rate. Spoofing attacks pose serious threats to traditional face recognition systems, which rely primarily on face detection and identification. The study's objective is to apply an anti-spoofing model in an MTCNN and FaceNet face recognition model utilizing the limited computing resources of Raspberry Pi. To achieve this, three specific objectives must be met; (1) To implement an MTCNN and FaceNet face recognition model that detects face input from a Raspberry camera (2) To implement a Color Texture and Corner Feature anti-spoofing detection model that identifies and prevents spoofing attacks in the face recognition model (3) Determine the accuracy of the facial recognition model with Anti-Spoofing Detection using the ratio of the False Reject Rate (FRR) and False Accept Rate (FAR). This study successfully implemented face image anti-spoofing, face detection, and identification on a Raspberry Pi system. At threshold 0.5, the anti-spoofing model achieves an accuracy of 94.94%, an FRR of 7.78%, and a low FAR of 1.47%. The face identification model gave the best results at a threshold of 0.1, with 100% Accuracy, 0% FAR, and a low FRR of 5.26%. The system allowed usability balancing with a very low FAR and FRR overall. Further improvements toward the real-time performance of the face recognition and anti-spoofing model are recommended for future studies.