DESIGN OF FACE RECOGNITION ANTI-SPOOFING MODEL BASED ON CONVOLUTIONAL NEURAL NETWORK USING FACENET TO ADDRESS VIDEO REPLAY ATTACKS

• Main Author: Fadillah Ramdhan, Lucky
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78321
• Institution: Institut Teknologi Bandung

Face-based biometric systems had received increasing attention because they had the richest and most easily accessible information in everyday life. However, the video replay attack had been a more challenging face presentation attack than other face presentation attacks because it could exhibit facial liveliness signals and was able to bypass face spoofing detectors. This research was developed using Design Research Methodology to design a Face Recognition Anti-Spoofing model to address video replay attacks. A FaceNet face recognition model with a Face Anti-Spoofing Convolutional Neural Network-based method was proposed and developed using feature matching techniques in an effort to improve authentication security. This research was conducted through field testing on 33 participants using their respective facial identities to measure the accuracy of the proposed model in the login scenario test and evaluated using System Engineering Principles and Practice. The test results using the feature matching technique showed that the accuracy of the model reached 97.57% and the F-score value was 96.22%. Therefore, the proposed model had proven to be effective in addressing video replay attacks through attack scenario testing.