Advanced Liveness Detection for Facial Recognition: Combating Fake Faces with Reflectivity Analysis

Facial recognition technology has become integral to daily life, facilitating processes such as online payments, smart door locks, and mobile phone unlocking, while enhancing security. However, its accuracy can be compromised by variations in illumination and action scale, making it vulnerable to high-fidelity fake face attacks. This project addresses these issues by identifying the differing reflectivity of human skin, video images, and fake face headgear under light. Utilizing spatio-temporal domain characteristics of facial key points, a novel face action liveness detection algorithm was developed. This algorithm integrates motion trajectory, historical data, and skin color information to improve liveness detection. The project processes facial video to extract reflective layers at key points, using this data to train a model capable of distinguishing real faces from fake ones, thereby improving the robustness of facial recognition systems.