Embedding Fast Temporal Information Model to Improve Face Anti-spoofing

Face anti-spoofing technology is a vital part of the face recognition system. For a quick response, many single-frame-based methods have been studied and made remarkable progress. However, some researchers improve performance by learning temporal features from video sequences without considering efficiency. Although the additional temporal features can improve face anti-spoofing, its computational efficiency is low. In this paper, we propose a fast temporal information model (Fast TIM) to learn temporal features. Fast TIM contains an efficient data dimensionality reduction method to retain temporal information and a lightweight network with 617 KB parameters to extract features. Fast TIM runs with 72 FPS real-time response and effectively improves the performance of the single-frame-based method. Experiments demonstrate that the proposed framework outperforms the state-of-the-art methods.