Abstract: This paper proposes a method for modeling the differences in illumination characteristics between living faces and fake faces using diffusion speed. Aiming at the need for face detection of mobile phone, according to the characteristics that the photo-reflection of the fake photo is more balanced and diffusion is slower than the living photo, a living detection method based on the diffusion speed model is proposed. The detection method obtains the diffusion velocity by introducing a total variation flow (TV), and uses the local velocity feature vector obtained by LSP coding (similar to LBP) as the input of the linear SVM classifier based on the obtained diffusion velocity map, and classifies the authenticity of the input image by classification. By designing multiple sets of comparative experiments, it is shown that the algorithm can obtain very good recognition results regardless of indoor environment or outdoor environment, multiple face poses and expressions, and various lighting conditions, and the LSP-based basic scheme is of high real-time and effective, can be deployed in a variety of mobile terminal devices, enable cross-platform one-button implant applications.

Key words: diffusion speed, total variation flow, local speed pattern, face liveness detection