Use FCN to Accurately Judge the Frontal Face or Side Face

Face-to-unlock is widely used in the fields of authentication, mobile payment and other information security. In the process of face-to-unlock, the detection for frontal face and side face can greatly avoid people's non-active intention triggering verification function and greatly improve the security of unlocking. Based on this reason, this paper establishes a set of model to judge frontal face and side face, by combining the latest FCN (Fully Convolutional Networks) algorithm with the geometric method creatively, to make the recognition rate for frontal and side face reach to 99.82%, far higher than by OpenCV traditional methods under the same data set. At the same time, in model training, this paper has improved training speed and model generalization ability by transfer learning technology.