Safe face recognition authentication system

The invention discloses a secure face recognition system for multifunctional interaction based on pytorch. The system has the functions of face recognition and authentication, face consistency detection, fine segmentation and face key point detection. And during face authentication, multi-azimuth face information input is adopted, so that the input accuracy is improved. And silence and dynamic living body detection is adopted during face login, so that cheating attacks such as secondary copying of a screen are effectively prevented. According to the face consistency detection method, a mobilenetv1 model is adopted, and Cross-Entry Loss and Triplet Loss are adopted as the total loss. Adversarial samples are added into the data set, and the probability of recognition errors is reduced. The fine segmentation is based on ResNet50, adopts a plurality of convolutions with different voidage rates to feel different feature maps, uses Semantic Estimation to locate the position of a portrait, exchanges the sequence of