首页 > 专利学习

基于改进的最近特征空间嵌入方法的人脸识别

硕士学位论文题目：基于改进的最近特征空间嵌入方法的人脸识别

英文并列题目：Face Recognition Based on

Improved Nearest Feature Space Embedding Method

研究生：杜弘彦

专业：软件工程

研究方向：现代软件工程技术

导师：王士同

指导小组成员：

学位授予日期：2018年6月

答辩委员会主席：陈秀宏

江南大学

地址：无锡市蠡湖大道1800号

二○一八年六月

摘要

人脸识别是时下很流行的重要研究方向，发展迅速。人脸识别技术利用了不同人的面部特征存在诸多差异这一现实，通过定位视频或图像里出现的人脸图像，加以处理后使之与已有的人脸库中的图像逐一匹配，从而完成身份识别。

历经半个世纪，人脸识别技术逐渐发达，加之人脸图像获取简便，成本低廉，容易为人们所接受，因此人脸识别系统被广泛应用于多种场合来提高身份鉴定的效率和准确性。本文介绍了一些经典的人脸识别算法，对它们的优缺点、发展方向进行了讨论。

传统的分类算法基于样本点之间的距离保留局部结构，没有充分利用类别信息，最近特征空间嵌入（NFSE）算法与拉普拉斯矩阵结合，引入特征空间的概念，更完整地保留了高维样本的拓扑结构和样本间线性关系。NFSE通过线性的欧氏距离度量来寻最近特征空间，导致类内离散度和类间离散度变化同步；测试时，最近邻分类器也使用欧氏距离度量，而忽视了高维空间样本间直线距离具有趋同性。这些都会降低识别率。另外，求每个样本的最近邻特征空间都要遍历所有类的做法使得训练时间较长。为此，本文在最近特征空间嵌入算法的基础上，结合非线性距离度量和样本夹角度量，对NFSE做了如下改进：1）提出了基于非线性距离和夹角组合的最近特征空间嵌入方法（NL-IANFSE）。

在训练阶段，该方法使用非线性距离度量方法选取近邻特征空间，使类内离散度的变化速度远小于类间离散度的变化速度，从而使转换空间中同类样本距离更小，不同类样本距离更大。在匹配阶段，使用结合夹角度量的最近邻分类器，充分利用样本相似性与样本间夹角的关系，更适合高维空间中样本分类。仿真实验表明，基于非线性距离和夹角组合的最近特征空间嵌入方法的性能总体上优于对比算法。

2）提出非线性距离的最近邻特征空间嵌入改进方法（NDNFSE），引入非线性距离公式选取最近邻特征空间，并使用结合夹角度量的最近邻分类器，提高了识别率；仅在样本的近邻类中选取最近邻特征空间，而不像NFSE遍历所有类，有效减少了训练时间。实验表明，NDNFSE的训练时间明显低于NFSE，识别率总体高于各对比算法。

关键词：人脸识别；非线性距离；夹角；最近特征空间嵌入；近邻类

Abstract

Face recognition is an important research field in pattern recognition and belongs to the category of

biometrics. Face recognition technology takes advantage of the fact that different human facial features have many differences, and completes identification by locating the face images appearing in the video or image, processing them and matching them with the images in the existing face library one by one.探针天线

After decades of development, face recognition technology has become increasingly mature. Facial images is convenient to obtain, low cost and easy to be accepted by people, so the face recognition system is widely used in many occasions to improve the efficiency and accuracy of identification. This article introduces some classic face recognition algorithms, and discusses their advantages and disadvantages, as well as their development direction.

1) In this study, nearest feature space embedding method based on the combination of nonlinear distance metric and included angle (NL-IANFSE) is proposed to overcome the above drawback. In the training phase, NL-IANFSE brings in nonlinear distance measure to make the rate of change of within-class scatter much slower than that of between-class scatter so that distances of samples within same class will be smaller and distances of samples belong to different classes will be larger in the transformed space. In the matching phase, NL-IANFSE uses the nearest neighbor classifier that combines Euclidean distance and included angle between two samples，taking the relationship

between similarity of samples and included angles of samples into account，which is more suited to classify in higher space. According to the experimental results, the proposed method outperforms the other algorithms for classification in high dimensional space.

2) Nearest feature space embedding method based on nonlinear distance metric (NDNFSE) was developed by using nonlinear distance formula to select the nearest feature spaces and using the nearest neighbor classifier combines Euclidean distance and included angle between two samples to improve the recognition rate. NDNFSE also sought every c lass’ nearest classes firstly, then only selected a sample’s nearest feature spaces within them to save the training time. According to the experimental results, NDNFSE outperforms comparison algorithms for classification as a whole, with a much shorter training time than that of NFSE.

正弦波信号发生器Keywords: face recognition; nonlinear distance；included angle；nearest feature space embedding; the nearest classes

II气门绞刀

摘要.......................................................................................................... II Abstract ........................................................................................................ II 目录......................................................................................................... III 第一章绪论 (1)

塑料玻璃滑道1.1 研究背景及意义 (1)

1.2 人脸识别概述 (2)

1.2.1 人脸识别的产生与发展 (2)

1.2.2 人脸识别技术的分类 (2)

1.2.3 人脸识别存在的问题 (3)痔疮仪

1.2.4 人脸识别步骤与特征提取 (3)

1.3 国内外研究动态 (4)

血竭提取物1.3.1 国内研究现状 (4)

1.3.2 国外研究现状 (4)

1.3.3 相关经典算法 (5)

1.4 常用人脸数据库 (6)

1.5 本文主要工作 (7)

1.6 本文章节安排 (8)

第二章流形学习算法与匹配规则 (9)

2.1 无监督降维算法 (9)

2.1.1 主成分分析 (9)

2.1.2 局部线性嵌入 (10)

2.1.3 局部保持投影 (11)

2.2 有监督降维算法 (12)

2.2.1 线性判别分析 (12)

2.2.2 改进的有监督局部线性嵌入算法 (13)

2.2.3 边界费舍尔分析 (13)

2.2.4 加强的边界费舍尔分析 (15)

2.2.5 最近特征空间嵌入方法 (15)

2.3 分类匹配规则 (16)

2.3.1 kNN算法 (16)

2.3.2 最近特征线匹配 (17)

2.3.3 结合夹角度量的最近邻分类器 (18)

2.4本章小结 (19)

III

本文发布于:2024-09-22 15:48:50，感谢您对本站的认可！

本文链接：https://www.17tex.com/tex/1/188044.html

上一篇：人脸识别外文翻译

下一篇：基于OPENCV具有眨眼侦测功能的人脸检测系统

标签：特征空间距离

留言与评论（共有 0 条评论）