Design And Implementation Of Face Recognition System In Uncontrolled Environment

As one of the hotspots in the fields of biometrics recognition and computer vision,face recognition in ideal environment has achieved good research results,but it is still a problem to perform accurate recognition in uncontrolled environments such as illumination and attitude changes.This paper studies the deep learning which has made breakthroughs in face recognition in recent years.Aiming at the problems of many methods such as weak feature recognition and poor face recognition performance under uncontrolled conditions,an improved face recognition algorithm based on deep learning is proposed,and a face recognition system for uncontrolled environments such as illumination and attitude changes is designed.The research work of this paper mainly includes:(1)The basic principles of deep learning and neural network are studied in detail firstly,and the mainstream face recognition method based on deep learning is analyzed and researched,which lays a theoretical foundation for follow-up work.(2)For unconstrained face images,face detection and feature point localization are realized by training multi-objective cascade neural network,and then face alignment is performed through affine transformation,and face image preprocessing is finally completed.FDDB,Wider Face and AFLW datasets' tests show that the face preprocessing algorithm has achieved good results.(3)Using the Softmax loss and the central loss on the pre-processed face data,the face recognition model is jointly supervised and trained on the deep neural network to optimize intra-class aggregation and inter-class dispersion.The tests show that the joint supervised algorithm achieves a higher recognition rate on both the LFW and YTF face authoritative data sets,and the accuracy is improved compared with the basic algorithm.(4)Combining the face detection and face recognition models,a face recognition system with face tracking and ambient illumination compensation is designed and implemented,and the function of the system is tested and verified on the self-built small data set.The experiments show that the system can perform certain compensation control on the uncontrolled environment in the field,which has good face discrimination ability and high accuracy.