基于深度学习的人脸识别算法浅析  

A Brief Analysis of Face Recognition Algorithm Based on Deep Learning

作　　者：康磊[1] 闫涛 KANG Lei;YAN Tao(School of Computer Science,Xi'an Shiyou University,Xi'an,Shanxi 710000)

机构地区：[1]西安石油大学计算机学院,陕西西安710000

出　　处：《长江信息通信》2022年第12期83-87,共5页Changjiang Information & Communications

摘　　要：相较于指纹识别、虹膜识别、声纹识别,人脸识别具有自然、便捷、体验友好的特征,成为大多数人认可的生物识别技术。近年来,随着GPU技术的成熟和数据集规模越来越大,让人脸识别技术的关注方向从基于手工特征的传统方法和传统的机器学习技术转移到使用大数据集训练的深度神经网络。现在,基于深度学习的人脸识别技术在人证比对、实名认证、人机交互、考勤、安防、美颜、趣味拍照、直播、微动作识别(疲劳驾驶、课堂听讲、罪犯审判)等领域得到了广泛的关注。文章首先简述人脸识别的发展历程,之后从深度学习方法、人脸数据集、网络结构、损失函数这四个方面,对目前流行的基于深度学习方法的人脸识别技术做一个较为详细的综述。Compared with fingerprint recognition,iris recognition,and voice recognition,face recognition has the characteristics of naturalness,convenience,and friendly experience,making it a biometric technology recognized by most people.In recent years,the maturity of GPU technology and the increasing size of data sets have shifted the focus of face recognition technology from traditional methods based on manual features and traditional machine learning techniques to deep neural networks trained using large data sets.Nowadays,face recognition technology based on deep learning has gained wide attention in the fields of human-ID matching,real-name authentication,human-computer interaction,attendance,security,beauty,fun photo-taking,live streaming,and micro-action recognition(fatigue driving,classroom listening,and criminal trial).This paper first briefly describes the development history of face recognition,and then gives a more detailed review of the currently popular face recognition techniques based on deep learning methods from four aspects:deep learning methods,face datasets,network structures,and loss functions.

关 键 词：深度学习 人脸识别 卷积神经网络 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]