Automatic Attendance System Using Silent Face Anti-Spoofing to Detect Spoof on Face Recognition

Some institutions or companies rely on attendance systems to assess the disciplinary performance of their employees so they can pay their employees's salaries fairly. One of the attendance systems that several institutions or companies often use is the facial recognition system, which is one of several biometric-based recognition systems. The Silent Face AntiSpoofing algorithm aims to address issues faced by institutions or companies regarding employee spoofing. This algorithm comprises two techniques, CNN (Convolutional Neural Network) and the Fourier transform. CNN is utilized for categorizing objects by their shape, like faces or other objects, whereas the Fourier transform is used for analyzing objects by their color wave frequency to differentiate between spoofed or real detected faces. The algorithm was trained using 15 million images to obtain an accurate model. It can distinguish between real faces and spoofed faces by analyzing the FT Loss results of real faces and spoofed faces. For some conditions based on light and detection distance, the algorithm is still confused when deciding which face is real and which is fake. The algorithm by data testing found its best performance when working outdoors at 09:00:00 UTC and 17:00:00 UTC, with a detection distance of 15 cm, and indoors with moderate lighting, with a detection distance of 30 cm. The key contribution of this research is the application of Silent Face Anti-Spoofing within a low-cost embedded system using Raspberry Pi 4, assessed in diverse lighting and distance conditions to enhance real-world attendance systems.