神经网络基于卷积神经网络的人脸识别研究与实现

电子科技大学

UNIVERSITY OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA硕士学位论文

MASTER THESIS

论文题目 基于卷积神经网络的人脸识别研究与实现

学科专业 软件工程

学 号 201321220122

作者姓名 万士宁指导教师 郝宗波 副教授

分类号 密级

UDC注1

学 位 论 文

基于卷积神经网络的人脸识别研究与实现

题名和副题名

万士宁

作者姓名

指导教师 郝宗波 副教授

电子科技大学 成 都

姓名、职称、单位名称

申请学位级别 硕士 学科专业 软件工程

提交论文日期2016.3.18 论文答辩日期 2016.4.19

学位授予单位和日期 电子科技大学 2016年6月

注1注明《国际十进分类法UDC》的类号。

Research and I mplementation of Face Recognition

Based on Convolution Neural Network

A Master Thesis Submitted to

University of Electronic Science and Technology of ChinaMajor: Software EngineeringAuthor: Shi ni ng WanSupervisor: Zongbo H aoSchool: School of I nformation and Software Engineering

独创性声明

本人声明所呈交的学位论文是本人在导师指导下进行的研究工作及取得的研究成果。据我所知除了文中特别加以标注和致谢的地方外论文中不包含其他人已经发表或撰写过的研究成果也不包含为获得电子科技大学或其它教育机构的学位或证书而使用过的材料。与我一同工作的同志对本研究所做的任何贡献均已在论文中作了明确的说明并表示谢意。

作者签名 日期 年 月 日

论文使用授权

本学位论文作者完全了解电子科技大学有关保留、使用学位论文的规定有权保留并向国家有关部门或机构送交论文的复印件和磁盘允许论文被查阅和借阅。本人授权电子科技大学可以将学位论文的全部或部分内容编入有关数据库进行检索可以采用影印、缩印或扫描等复制手段保存、汇编学位论文。

保密的学位论文在解密后应遵守此规定

作者签名 导师签名

日期 年 月 日

摘要

摘 要

如今随着计算机视觉的相关理论与应用研究的快速发展计算机视觉技术在日常生活应用中的优越性日益突显出来。本文主要研究了深度学习方法中的卷积神经网络模型在自然场景下人脸识别领域的应用。深度卷积神经网络模型相比较于传统的人脸识别的方法不需要人工进行复杂而耗时的特征提取算法设计只需要设计一个有效的神经网络模型然后在大量的训练样本上进行端到端的简单、高效的训练就能获得不错的分类准确率。该方法的性能和效果主要取决于网络结构的设计 因此本文研究重点在于构建一个合理的网络模型结构并采取一些相关技术保证其在训练集上能够稳定地、快速地收敛而且还要最终获得良好的分类准确率。

本文主要内容包括

1论文中对卷积神经网络的基础理论知识进行了归纳总结。卷积神经网络发展于传统的神经网络本文先从早期的传统神经网络中的网络结构、梯度下降、BP算法(Error Back Propagation)进行了阐述。然后过渡到卷积神经网络的理论基础并对其中的一些关键的非线性计算的卷积层、下采样层等进行了阐述。最后通过经典的卷积神经网络LeNet-5的例子说明了卷积神经网络模型的一般整体结构。

2通过合理的减少原V GG卷积神经网络训练参数得到了改进的Li ghtenedV GG网络模型并使用比随机初始化更好地参数初始化方法来缩减模型的收敛时间最终该新模型不仅解决了原V GG模型对硬件要求高、训练困难等方面的问题而且成功的应用于自然环境下的人脸识别并在严格预处理后的LFWLabe l e dFaces in the Wild人脸数据库上进行实验获得了94%的准确率。然后在这个模型之后增加了一个Siamese神经网络模型提升了该网络对较为复杂的人脸图片的特征提取能力。论文也对该Siamese模型进行详细的介绍和分析。

3论文采用一种新的残差学习思想来构建了一个全新的应用于人脸识别领域的Residual网络模型。该模型深度达到了34层采用了新的参数初始化方式来解决深度网络的收敛难问题并使用了批度归一化Batch Normalization技术增加了模型的稳定性。通过在LFW人脸数据库上面进行实验取得了比LightenedV GG模型更好的96%左右的准确率。

4最后将上述的模型算法应用于实际场景中实现了一个基于实时监控视频的人脸识别系统。对系统各个模块的功能和流程进行详细介绍并在自建的

I

摘要

人脸数据库上进行了测试达到了93%的准确度。该系统验证了本文方法的有效性达到了在监控视频中进行人脸识别的应用要求。

关键词人脸识别卷积神经网络 LFW数据库 Siamese模型特征提取

II

ABSTRACT

ABSTRACT

Nowadays,with the rapid development of the related theories and applications ofcomputer vision, the superiority of the application of computer vision technology indaily life is becoming more and more important. This thesis mainly studies theapplication of the convolution neural network model which is belong to deep learningmethod in the field of human face recognition in natural scenes.Compared with thetraditional face recognition method, the deep Convolutional Neural Network model(CNN)does not need to design the feature extraction algorithm,which is complex andtime-consuming, itjust have to design an effective neural network model,and the modellearn from a large number of training samples by an end to end training, then thismethod can reach an good classification accuracy.The performance and effectiveness ofthe method are mainly determined by the design of the model structure,so the key pointof this thesis is to design a reasonable neural network model, and some relatedtechnologies is also applied in the model to ensure that the model can converge on thetraining set quickly and stably.

The main contents of this thesis include:

(1) In this thesis, some basic theories of the convolutional neural network aresummarized. Convolutional neural network developed from the traditional neuralnetwork,so the network structure of traditional neural network,gradient descent methodand BP algorithm(Error Back Propagation Algorithm) are described.And thentransition to the description of the related theories of convolutional neural network,suchas convolutional layer, pooling layer etc. Finally, this thesis illustrates the generalstructure of convolution neural network model by introducing the classic LeNet-5network.

(2)By reducing the number of parameters in the raw VGG convolutional neuralnetwork reasonably,an improved Lightened VGG network model has been designed,and a new parameter initialization method is applied in this model,which is better thanrandomly parameter initialization method, to reduce the time of model convergence.Atlast, this new model not only solves some issues which had occurred in the originalmodel,such as higher-quality hardware requirements, the difficult of training,and so on,but also is successfully applied to face recognition in natural scene,which reached 94%

III

ABSTRACT

accuracy rate on the strictly pre-processed LFW(Labeled Faces in the Wild)dataset.Then,to further improve the ability of the model to extract the features of more compleximages,a Siamese model is used and illustrated in detail.

(3) In this thesis, a residual convolutional neural network also is designed byapplying a new residual learning theory.The layers of this model reached to 34.Tosolve the difficult of convergence in this model,a new parameter initialization methodis used,and Batch Normalization technique is applied to make the model more stable.According to the result on LFW, the accuracy rate of this model can reached 96%,which is better than the Lightened VGG model.

(4)Finally,a face recognition system based on the Real-Time surveillance video inreal scenario is implemented by applying the models mentioned before.The functionand process of each module in the system are illustrated in detail,and the accuracy ofthe test,which is carried out on a self-built face database, is 93%.The system verifiesthe effectiveness of this method,and it can meet the requirements of face recognitionapplications in the surveillance video.

Keywords: face recognition, convolutional neural network,LFW database, Siamesemodel,feature extraction

IV

目录

目录

第一章绪论...................................................................................................................1

1.1研究背景与意义................................................................................................1

1.2国内外发展现状................................................................................................1

1.3人脸识别技术的发展........................................................................................3

1.3.1传统人脸识别的发展..................................................................................3

1.3.2基于深度学习的人脸识别技术发展..........................................................6

1.4本文主要工作....................................................................................................7

1.5本论文的结构安排............................................................................................8

第二章卷积神经网络的理论基础.................................................................................9

2.1神经网络的理论基础........................................................................................9

2.1.1前馈神经网络的结构..................................................................................9

2.1.2梯度下降....................................................................................................10

2.1.3误差反向传播Error Back Propagation算法......................................12

2.2卷积神经网络结构的组成..............................................................................15

2.2.1各种线性计算层........................................................................................15

2.2.2激活函数....................................................................................................17

2.2.3网络整体架构............................................................................................18

2.3本章小结..........................................................................................................20

第三章Lightened VGG卷积神经网络.......................................................................21

3.1 ReLU(Rectified Linear Units)...........................................................................21

3.2 Dropout..............................................................................................................23

3.3 Lightened VGG卷积神经网络模型................................................................23

3.3.1原V GG网络结构简析.............................................................................24

3.3.2 Lightened VGG神经网络模型..................................................................26

3.4 Lightened VGG模型的训练与实验分析........................................................30

3.4.1图片数据库与预处理................................................................................30

3.4.2 Lightened VGG网络模型的训练与实验分析..........................................32

3.5 Siamese网络模型.............................................................................................39

3.5.1 Contrastive Loss Function...........................................................................39

V