上传用户：fosyhffmjj资料价格：5财富值&&『』文档下载 ：『』&&『』学位专业：&关 键 词 ：&&&&&权力声明：若本站收录的文献无意侵犯了您的著作版权，请点击。摘要:（摘要内容经过系统自动伪原创处理以避免复制，下载原文正常，内容请直接查看目录。）跟着盘算机技巧的飞速成长和社会公共平安须要的日趋增加，身份辨认遭到极年夜的看重，作为典范的生物特点辨认的人脸辨认也逐步成为研讨热门之一。虽然线性质空间办法在人脸辨认中曾经获得了极年夜的胜利，然则研讨注解人脸图象极可能是散布在一个嵌入到高维图象空间的低维非线性质流形上，且人脸辨认常常是高维小样本成绩，在小的样本集练习获得的子空间分类器后果不是很好。本文以流形进修为基本，对非参数核流形进修降维办法、半监视流形正则化分类办法停止深刻的研讨。本文重要停止了以下研讨任务：(1)研讨基于非参数核谱回归的降维办法经由过程对流形进修相干实际和技巧在特点提取中的研讨，针对图嵌入核扩大办法中的核函数的选择和结构影响着办法的机能，且流形进修降维平日须要触及盘算时光、空间庞杂度高的浓密矩阵的特点分化成绩。本文基于流形进修图嵌入降维框架，引入非参数核进修和谱回归办法，提出基于非参数核谱回归的降维办法，有用防止了针对特定成绩核函数选择和结构困难，具有高效性和可扩大性。(2)研讨基于半监视流形正则化的极速进修机办法针对基于核的半监视进修办法具有高盘算庞杂度，联合流形正则化和成对束缚信息，和疾速进修才能的极速进修机办法，提出一种新的基于半监视流形正则化的极速进修机算法，供给了传统的半监视办法的一种近似办法，不只实用到半监视的情形，并且其决议计划函数的情势实用于年夜范围的进修义务，在真实数据集上的试验成果验证了所提出算法的有用性。(3)设计并完成基于流形进修的人脸辨认原型体系经由过程改良人脸辨认进程中的人脸数据降维办法和人脸辨认分类办法，设计并完成了基于非参数核谱回归的降维算法和半监视流形正则化的极速进修机的原型体系，应用ORL、Yale及人脸数据库作为试验数据完成了软件体系设计。Abstract:Along with the computer technology rapid development and social public safety needs is increasing day by day, the identification was greatly valued, as a model of biological characteristics recognition of face identification has gradually become one of the hot research. Although the properties of space line method in face recognition once won a huge victory, however research notes face image very likely is spread in an embedded to the high dimensional image space to lower dimensional non linear manifold nature, and face recognition is often high-dimensional small sample performance, in the small sample set exercises the subspace classifier consequences are not very good. This paper studies the basic manifold, manifold learning of non parametric kernel dimensionality reduction methods, semi supervised Manifold Regularization classification method to stop deep discussion. In this paper, an important stop the following research work: (1) research based on nonparametric regression dimensionality reduction method of nuclear spectrum through of manifold learning coherent theory and techniques in the feature extraction of research, for expanding the graph embedding nuclear way of kernel function selection and structure affects function approach, and manifold learning dimension reduction usually need to touch the computing time and space complexity high dense matrix differentiation characteristics of achievement. The based on manifold learning graph embedding dimensionality reduction framework, the introduction of non parametric kernel learning and spectral regression method, is proposed based on non parameter nuclear spectrum regression of dimensionality reduction methods, effectively prevented the specific results the choice of kernel function and structure difficult, with efficiency and scalability. (2) the research based on the semi supervised Manifold Regularization speed of machine learning approach for based on nuclear semi supervised learning methods with high computing complexity, combined with Manifold Regularization and pairwise constraint information, and learn fast to the speed of machine learning approach proposed a new semi supervised manifold regularized extreme learning machine algorithm based on supply the traditional semi supervised approach, an approximate method, not only practical to semi supervised and practical situation for the decision function to study obligations of large-scale, in real data sets of test results verify the algorithm proposed in this paper is useful. (3) designed and completed a manifold learning based face identification prototype system by improving their face identification in the process of face data descending dimension method and face identification classification method, designed and completed the prototype system based on the non parametric spectrum kernel regression algorithm for dimensionality reduction and semi supervised Manifold Regularization extreme learning machine, using the ORL and yale face database as data completed the design of software system.目录:致谢4-5摘要5-6Abstract6-7目录8-10Contents10-12图清单12-14表清单14-151 绪论15-21&&&&1.1 人脸识别背景与意义15-16&&&&1.2 人脸识别的研究现状16-18&&&&1.3 本文的主要研究内容18-19&&&&1.4 本文结构及章节安排19-212 相关理论与技术21-31&&&&2.1 引言21&&&&2.2 流形学习21-23&&&&2.3 几种代表性的流形学习算法23-27&&&&2.4 极速学习机27-30&&&&2.5 本章小结30-313 基于非参数核谱回归的降维算法31-42&&&&3.1 引言31&&&&3.2 图嵌入降维算法31-33&&&&3.3 非参数核学习算法33-35&&&&3.4 基于非参数核谱回归的降维算法35-38&&&&3.5 实验结果与分析38-40&&&&3.6 本章小结40-424 基于半监督流形正则化的极速学习机算法42-54&&&&4.1 引言42&&&&4.2 半监督学习42-46&&&&4.3 流形假设和成对约束信息46-47&&&&4.4 基于半监督流形正则化的极速学习机算法47-49&&&&4.5 实验结果与分析49-53&&&&4.6 本章小结53-545 原型系统设计与实现54-60&&&&5.1 引言54&&&&5.2 系统设计与实现54-57&&&&5.3 系统功能演示57-59&&&&5.4 本章小结59-606 总结与展望60-63&&&&6.1 本文总结60&&&&6.2 进一步的研究工作60-63参考文献63-67作者简历67-69学位论文数据集69分享到：相关文献|