基于神经网络的人脸识别研究综述

A Survey of Research on Face Recognition Using Convolutional Neural Network

人脸识别是利用计算机分析人脸视频或者图像,从中提取有效识别信息,并最终判断人脸对象身份的技术,是图像分析领域最重要的应用之一。随着神经网络,尤其是卷积神经网络(CNN)的快速发展,学者和工业界发现CNN在处理图像以及提取图像特征方面相比于传统机器学习技术,表现出更高的准确率和更好的鲁棒性。这也让CNN广泛运用于当代的人脸识别技术。本文将介绍人脸识别技术的一般步骤,从人脸检测、人脸对齐到人脸表征和人脸匹配,对每一步骤都进行详细介绍和文献综述,然后介绍卷积神经网络,并对影响CNN准确率的3大影响因素(数据集、模型架构、损失函数)的相关经典研究进行阐述,最后对未来的研究方向进行预测,期望继续推动本领域的研究与发展。

Face recognition is a technology that uses computers to analyze face videos or images,extracts effectively recognition information from images,and finally determines the identity of a face object.It is one of the most important applications in the field of image analysis.With the rapid development of neural networks,especially convolutional neural networks(CNN),scholars and industry stakeholders have found that CNN has higher accuracy and better robustness than traditional machine learning techniques in terms of processing images and extracting image features.This allows CNN to be widely used in contemporary face recognition technologies.This article will introduce the general steps of face recognition technology,from face detection and face alignment to face representation and face matching.Each step will be introduced in detail and related literature will be reviewed,and then the convolutional neural network will be introduced,and then classic studies on the three major influencing factors of CNN’s accuracy(datasets,model architectures,and loss functions)are elaborated,and finally three predictions are made for the future research directions,hoping to promote research and development in this field.