End-to-End Deep Convolutional Neural Network For Face Recognition

Face recognition is the problem of identifying a specific individual,rather than merely detecting the presence of a human face.It is widely used in the public security,finance security and commercial domain.Because of this,face recognition is a core problem and popular research topic in computer vision.The recent face recognition methods are based on deep learning,and have made a great progress,even beated human beings on the benchmark LFW.In order to fully exploit the ability of deep convolutional neural network,we employ an end-to-end model and propose a novel and more appropriate loss function for face recog-nition.Our model trained with public available datasets and achieved accuracy 99.15%on the benchmark LFW with only single model and with alignment.Even without alignment,we also achieved accuracy 99.08%,which exceeded the best model Google FaceNet under the same circumstance with an accuracy 98.87%.After simple model ensemble,we achieved 99.33%with alignment and 99.28%without alignment,and both exceeded human perfor-mance 99.20%.Finally,we try to analyse and understand face recognition system with visualization.