Research And System Inplementation Of Face Recognition Based On Deep Learning Under Unrestricted Conditions

With the acceleration of the construction of smart cities,technologies such as big data and artificial intelligence are becoming more and more mature.Face recognition technology based on the above theory has been rapidly developed.Automatic face recognition as a biometric authentication technology for identification has been widely used in military,public safety and daily life.Most of the existing face recognition products are suitable for normal postures and high-definition static face images that can be controlled under controlled conditions.For those faces taken under unrestricted conditions,such as face-mount camera on the road surface and video face images taken by ordinary pedestrian cameras,it usually contains more time-related information and multi-view information,but the faces are often blocked,the offset angle is too large,the pixels are low and the illumination is dark,etc.,These situations will increase the difficulty of recognition.Based on the above reasons,this paper intends to study how to improve the automatic face recognition effect in non-restricted video scenes,while taking into account the performance of the algorithm.This paper proposes a new deep network structure for automatic face detection of images under unrestricted conditions,which can realize automatic detection of faces at various scales.At the same time,considering the reasons of data confidentiality,the publicly released video face card has very few data.In order to solve the modeling problem of this small sample,this paper proposes a face recognition network that uses small-scale data for training.The training method is trained using the weighted decomposition training method,and the model with the same effect as the large data set training model is obtained,which improves the accuracy of face recognition under unrestricted conditions.The experimental results show that the two algorithms proposed in this paper can achieve automatic detection and automatic recognition of faces under unrestricted conditions,different illumination,occlusion and so on.