QQ客服 官方微博 反馈留言

题名	静态场景下人脸的检测与识别
作者	刘威
学位类别	工学硕士
答辩日期	2004-07-01
授予单位	中国科学院研究生院
授予地点	中国科学院自动化研究所
导师	王蕴红 ; 谭铁牛
关键词	人脸检测 人脸识别 了空间分析 零空间 核Fisher判别分析 类内空间 face detection face recognition subspace analysis null space kernel fisher discriminant analysis intra-class space manifold lear
其他题名	Face Detection and Recognition in a Still Background
学位专业	模式识别与智能系统
中文摘要	人脸检测和识别的研究目的是使计算机具有类似人类的脸像识别能力。它在身份鉴 别、人机交互、图像检索和视觉监控等众多领域有着重要的应用。尽管近三十年里，人脸 检测和识别受到了国内外学术界和企业界的广泛关注，但作为模式识别和计算机视觉领域 中的一个研究热点和难点，人脸检测和识别仍有很多理论与技术问题需待解决。有鉴于 此，本文围绕静态图像中的人脸检测和识别研究中存在的一些重要问题开展了深入研究， 主要工作可以归纳如下： （1）完整地研究了基于扩展的Haar特征与Adaboost分类学习算法的实时在线检测技 术，成功地实现了整个人脸检测系统。 （2）针对小样本情况下LDA出现的问题，进行零空间（null space）的分析，提出零空 间线性判别分析（NLDA）方法，实验证明NLDA是小样本情况下LDA最好的扩 展。 （3）结合零空间方法与核Fisher判别分析，提出新的非线性人脸识别方法：基于零空间 的核Fisher判别分析（Null Space-based Kernel Fisher Discriminant Analysis or NKFDA）。NKFDA方法用Cosine核函数来代替多项式核函数，基于最小的重构 误差构造核映射和核样本集，在核样本集上执行简化了的NLDA算法，最后输出一 个全局的非线性维数降低映射。NKFDA方法仅仅执行一次特征分解，实验证明在 训练样本数量很多（大于300个）的情况下，提出的NKFDA在识别率上优于典型 的LDA方法。 （4）利用虚拟样本这一重要思想，提出了一个新的分类方法：最近类内空（nearestintra-class space oi NICS）。每类的人脸模式分布由类内空间（intra-class space or ICS）表示和描述，它捕捉和覆盖各式各样的类内变化，所有的虚拟样本都 在类内空间被翻建；然后基于子空间方法定义了两个距离量度：子空间外距离 （distance-from-subspace）和子空间内距离（distance-in-subspace），二者的加 权就是测试样本到类内空间的距离量度，分类规则将基于最短的加权距离判定。 （5）使用流形学习技术（manifold learning）建立虚拟的局部流形结构，以局部 流形为单位提出新的分类方法：局部流形匹配（local manifold matching or LMM）。LMM方法假设每个特征点和它的近邻存在于一个虚拟的线性局部流 形中，使用局部流形学习技术于虚拟的流形结构，学习出所有的局部流形；然后定 义测试样本到局部流形的距离量度，最匹配的局部流形将基于最短的流形
英文摘要	Research on face detection and recognition aims at enabling machines to possess face recognition ability similar to that of human beings. It has essential applications in the domain of personal identification, human-computer interface, image retrieval, visual surveillance, etc. Although the topic has received much attention in pattern recognition and computer vision community, many open problems remain to be resolved. In this thesis, we describe several novel algorithms in an attempt to solve some of the problems. The main contributions of this thesis include following issues: (I) We develop a demo platform to implement a static detector for detecting face regions from arbitrary BMP or JPG files on a computer with PIV 1.7G CPU and Windows operating system. Our system uses Haar-like features, which are computed by integral image rapidly, and applies Adaboost algorithm to select important detecting features. Then, a cascaded classifier combining many classifiers is trained and allows background regions of the image to be quickly discarded. Not only does the detector ensure a enough intriguing detection rate, but it also persist a substantial small false positive. (2) We propose a modified LDA method in case of the small sample size problem(SSSP), it is called NLDA. The null space of the within-class scatter matrix is found to express most discriminative information for SSSP. Our method takes full advantage of the null space while the other methods re- move the null space. It proves to be optimal in performance. (3) We incorporate kernel technique into discriminant analysis in the null space and derive null space-based kernel fisher discriminant analysis(NKFDA). Firstly, all samples are mapped to the kernel space through a better kernel function, called Cosine kernel, which is proposed to increase the discriminating capability of the original polynomial kernel function. Secondly, a truncated NLDA is employed. The novel approach only requires one eigenvalue analysis and is also applicable to the large sample size problem. Experiments are carried out on different face data sets to demonstrate the effectiveness of the proposed methods. (4) we propose a novel classification method, called nearest intra-class space (NICS), for face recognition. The distribution of face patterns of each person is represented by the intra-class space to capture all intra-class variations. Then, a regular principal subspace is derived from each intra-class space using principal component analysis. The classification is based on the nearest weighted distance, combining distance-from-subspace and distancein-subspace, between the query face and each intra-class subspace. Experimental results show that the NICS classifier outperforms other classifiers in terms of recognition performance. (5) we present
语种	中文
其他标识符	764
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/6801]
专题	毕业生_硕士学位论文
推荐引用方式 GB/T 7714	刘威. 静态场景下人脸的检测与识别[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2004.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。