基于深度卷积神经网络的人脸识别技术综述  

A SURVEY OF FACE RECOGNITION TECHNOLOGY BASED ON DEEP CONVOLUTIONAL NEURAL NETWORKS

作　　者：景晨凯 宋涛[1] 庄雷[1] 刘刚[2] 王乐[2] 刘凯伦 

机构地区：[1]郑州大学信息工程学院,河南郑州450001 [2]河南省招生办公室,河南郑州450046

出　　处：《计算机应用与软件》2018年第1期223-231,共9页Computer Applications and Software

基　　金：国家自然科学基金项目(61379079);河南省国际合作项目(152102410021)

摘　　要：人脸识别是计算机视觉的重要应用之一,广义的人脸识别包含图像采集、人脸检测、人脸对齐、特征表示等过程。人脸识别的发展史主要是人脸特征表示方法的变迁史。针对特征的表示方法,从人脸识别技术的发展历史、研究现状和未来发展三个方面进行综述:分阶段对传统的几类经典的人脸识别算法进行回顾和总结;以深度学习算法的诞生过程为切入点,重点分析了在人脸识别中取得突破性进展的深度卷积神经网络DCNN(deep convolutional neural networks)的技术思想和关键问题;针对人脸识别和深度学习算法的重大挑战,展望了未来可能存在的发展方向。Face recognition is one of the important applications of computer vision. Generalized face recognition includes image acquisition, face detection, face development history of face recognition is mainly alignment, and feature representation and so on. However, the the history of the change of face feature representation method and summarizes the three aspects of the development history, research status and future development of face recognition technology. Firstly, reviews and summarizes several kinds of classical phases of traditional face recognition algorithm. Secondly, based on the process of the depth learning algorithm, the technical ideas and key problems of deep convolutional neural networks （DCNN）, which are the breakthrough progress in face recognition, are analyzed emphatically. Based on this, the paper finally talks about the prospect of face recognition in the direction of development may exist in the future under the challenge of face recognition and deep learning algorithms.

关 键 词：人脸识别 特征表示 深度学习 深度卷积神经网络 

分 类 号：N391.41[自然科学总论]