Face Detection And Face Recognition Under Unconstrained Environment

With the development of computer vision,face recognition have become significant research directions.The technology of face recognition generally consists of three parts:face detection,face alignment and face recognition.Aiming at face detection and face recognition under unconstrained environment,we propose several innovations for face detection and face recognition,respectively.And we also implement a demo for face recognition with practical significance.Details of main research works are as follows:For face detection,(1)we design a one-stage face detector based on Region Proposal Network(RPN).The one-stage face detector has a relatively high accuracy,with smaller model size and faster face detection speed;(2)We propose a novel training strategy,Precise Box Score(PBS),which can extract more information from face detection dataset.The performance of face detection model can be boosted significantly without changing the train set and network structure;(3)We propose a simple but effective model compression method(SEMCM),which uses a big model with higher accuracy to train a small model with lower accuracy.Experimental results show the accuracy of small model can be raised significantly.For face recognition,we propose an auxiliary loss function,contrastive-center loss,based on metric learning.The proposed supervision signal can boost the accuracy of face recognition model effectively,without influencing network testing time.What is more,we also explore the influence of similarity search on face recognition.For the application of face technology,we design and implement face searching system and face monitor system,respectively.