LA4OX - A Live Authentication for Online Exam

Online learning systems, especially remote learner assessment systems, are experiencing difficulties with verifying the identity of the learners. The username-password approach faces two main issues when applied to online examination systems: (1) It is difficult to remember strong passwords (2) It is difficult to verify the student who is assessing. Live authentication utilizing biometrics can be used to solve these two issues. However, it is challenging for a single biometric system to maintain stability and prevent spoofing. Many studies have shown that multibiometrics systems outperform unibiometric systems, but they have typically been implemented using simple fusion methods. In this work, we propose a feature-level and score-level fusion method utilizing multi-head attention module. Separately, facial features and voice features are extracted, then they are fused by the multi-head attention module. Moreover, two protocols for live authentication are presented in this paper. The proposed model is tested on a popular dataset and a real-world dataset. On VoxCeleb 1 dataset, the proposed method outperforms individual biometrics in terms of validation results, demonstrating accuracy rates of 98.66% for voice-only, 95.72% for face-only, and 99.75% for our model.