本文方法实验结果课件.pptx
- 【下载声明】
1. 本站全部试题类文档,若标题没写含答案,则无答案;标题注明含答案的文档,主观题也可能无答案。请谨慎下单,一旦售出,不予退换。
2. 本站全部PPT文档均不含视频和音频,PPT中出现的音频或视频标识(或文字)仅表示流程,实际无音频或视频文件。请谨慎下单,一旦售出,不予退换。
3. 本页资料《本文方法实验结果课件.pptx》由用户(晟晟文业)主动上传,其收益全归该用户。163文库仅提供信息存储空间,仅对该用户上传内容的表现方式做保护处理,对上传内容本身不做任何修改或编辑。 若此文所含内容侵犯了您的版权或隐私,请立即通知163文库(点击联系客服),我们立即给予删除!
4. 请根据预览情况,自愿下载本文。本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
5. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007及以上版本和PDF阅读器,压缩文件请下载最新的WinRAR软件解压。
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- 本文 方法 实验 结果 课件
- 资源描述:
-
1、Eigenfeature Regularization and Extraction in Face Recognition作者作者:讲解人:讲解人:1谢谢观赏2019-5-9提纲l文章信息文章信息l背景介绍背景介绍l本文方法本文方法l实验结果实验结果2谢谢观赏2019-5-9提纲l文章信息文章信息作者信息文章信息摘要l背景介绍背景介绍l本文方法本文方法l实验结果实验结果3谢谢观赏2019-5-9第一作者l Xudong Jiang Asst Professor Director center for Information Security(CIS)School of Electrical
2、and Electronic Engineering Nanyang Technological University http:/wwwl 简历简历 B.Eng.And M.Eng.University of Electronic Science and Technology of China(电子科技大电子科技大学,成都学,成都),1983,1986.Lecturer University of Electronic Science and Technology of China,19861993 Scientific Assistant Helmut Schmidt University
3、 Hamburg,19931997 Ph.D.Helmut Schmidt University Hamburg,Germany,electrical and electronic engineering,1997.4谢谢观赏2019-5-9第一作者Research Fellow Centre for Signal Process(CSP),Nanyang Technological University,Singapore,19982002.where he developed a fingerprint verification algorithm that achieved the fa
4、stest and the second most accurate fingerprint verification in the International Fingerprint Verification Competition(FVC2000).Lead Scientist and Head Biometrics Lab at the Institute for Infocomm Research,A*Star(Agent for Science,Technology and Rearch),Singapore,20022004Adjunct Assistant ProfessorCI
5、S,Nanyang Technological University,20022004Assistant ProfessorCIS,Nanyang Technological University,2004now5谢谢观赏2019-5-9第一作者lPublication:X.D.Jiang,“Asymmetric Principal Component and Discriminant Analyses for Pattern Classification,”IEEE TPAMI,Vol.31,No.5,pp.931-937,May,2009.X.D.Jiang,B.Mandal and A.
6、Kot,“Eigenfeature Regularization and Extraction in Face Recognition”IEEE TPAMI,Vol.30,No.3,pp.383-394,March,2008.X.D.Jiang,M.Liu and A.Kot,“Fingerprint Retrieval for Identification,”IEEE Transactions on Information Forensics and Security,Vol.1,No.4,pp.532-542,December 2006.X.D.Jiang,“On Orientation
7、and Anisotropy Estimation for Online Fingerprint Authentication,”IEEE TSP,Vol.53,No.10,pp.4038-4049,October 2005.K.Toh,X.D.Jiang and W.Yau,“Exploiting Global and Local Decisions for Multi-Modal Biometrics Verification,”IEEE TSP,Vol.52,No.10,pp.3059-3072,October 2004.X.D.Jiang and W.Ser,“Online Finge
8、rprint Template Improvement”,IEEE TPAMI,vol.24,no.8,pp.1121-1126,August 2002.Many ICIP,ICPR papers6谢谢观赏2019-5-9lCurrent Research Areas:Statistical Pattern RecognitionComputer VisionMachine LearningImage and Signal ProcessingBiometricsFace RecognitionFingerprint Recognition7谢谢观赏2019-5-9第二作者l Bappadit
9、ya Mandal research fellowInstitute for Infocomm Research,A*Star,department of Computer Vision and Image Understanding,2008now Ph.D.of Xudong Jiang and Alex kot,in CIS,20042008 B.Techin Electrical Engineering,India Institute of Technology,Roorkee,India,19992003“If we knew what it was we were doing,it
10、 would not be called research,would it?”-Albert Einstein http:/www8谢谢观赏2019-5-9第二作者l Publication:B.Mandal,X.D.Jiang and A.Kot,“Face Verification Using Modeled Eigenspectrum,”The Open Artificial Intelligence Journal,Bentham Open,19th May 2008 B.Mandal,X.D.Jiang and A.Kot,“Verification of Human Faces
11、Using Predicted Eigenvalues,”International Conference on Pattern Recognition(ICPR 2008),Tempa,Florida,USA,8-11 Dec 2008(oral presentation).Received the Best Biometrics Student Paper Award.B.Mandal,X.D.Jiang and A.Kot,“Dimensionality Reduction in Subspace Face Recognition,”IEEE Sixth International Co
12、nference on Information,Communications and Signal Processing(ICICS 2007),pp.1-5,Singapore,10-13 December 2007 B.Mandal,X.D.Jiang and A.Kot,“Kernel Fisher Discriminant Analysis in Full Eigenspace,”International Conference on Image Processing,Computer Vision,and Pattern Recognition(IPCV 2007),Las Vega
13、s,Nevada,USA,pp.235-241,25-28 June 2007.B.Mandal,X.D.Jiang and A.Kot,“Multi-scale feature extraction for face recognition,”IEEE International Conference on Industrial Electronics and Applications(ICIEA 2006),Singapore,pp.1-6,24-26 May 2006(invited paper).9谢谢观赏2019-5-9第三作者l Alex Kot Prof.of Nanyang T
14、echnological Univeristy(NTU),Singapore since 1991.Associate Editor for IEEE Trans.On Signal Processing,20002003 Associate Editor,IEEE Trans.On Circuits and Systems Part II,20042006,Part I 20052007.IEEE Distinguished Lecture in 2005 and 2006 Fellow of IEEE and IES(Industrial Electronics Society)http:
15、/www10谢谢观赏2019-5-9l Researches:Information TechnologySecurity in Black and WhiteSteganalysis&Image ForensicsBiometrics:Signature,Face and FingerprintSignal Processing for CommunicationsRejection of Interference in Spread Spectrum Systems Using Signal Processing Techniques Space-time modulation文章多是和别
16、人合作的11谢谢观赏2019-5-9文章出处l题目题目:Eigenfeature Reuglarization and Extraction in Face Recognitionl出处:出处:IEEE TPAMI,vol.30,no.3l时间:时间:2008.3l相关文献:相关文献:12谢谢观赏2019-5-9AbstractlThis work proposes a subspace approach that regularizes and extracts eigenfeatures from the face image.lEigenspace of the within-class
17、 scatter matrix is decomposed into three subspaces:a reliable subspace spanned mainly by the facial variation,an unstable subspace due to noise and finite number of training samples,and a null subspace.lEigenfeatures are regularized differently in these three subspaces based on an eigenspectrum mode
18、l to alleviate problems of instability,overfitting,or poor generalization.lThis also enables the discriminant evaluation performed in the whole space.Feature extraction or dimensionality reduction occurs only at the final stage after the discriminant assessment.These efforts facilitate a discriminat
19、ive and a stable low-dimensional feature representation of the face image.lExperiments comparing the proposed approach with some other popular subspace methods on the FERET,ORL,AR,and GT databases show that our method consistently outperforms others.13谢谢观赏2019-5-9摘要l 本文提出了一种从人脸图像中提取和正则化本征特征的子空间本文提出了
20、一种从人脸图像中提取和正则化本征特征的子空间方法。方法。l 类内散度矩阵的特征空间被分解为三个子空间类内散度矩阵的特征空间被分解为三个子空间:主要有面部变:主要有面部变化张成的可靠的子空间,由噪声和有限样本导致的不稳定子空化张成的可靠的子空间,由噪声和有限样本导致的不稳定子空间,以及零空间。间,以及零空间。l 基于特征谱模型,基于特征谱模型,分别在这三个子空间中对本征特征进行不同分别在这三个子空间中对本征特征进行不同的正则化的正则化,从而减轻了不稳定、过拟合、推广能力差的问题。,从而减轻了不稳定、过拟合、推广能力差的问题。也使得也使得判别估计在整个空间进行判别估计在整个空间进行。特征提取和降维
21、只是在最后。特征提取和降维只是在最后一个阶段进行,而这是判别估计之后的。这些工作使得人脸图一个阶段进行,而这是判别估计之后的。这些工作使得人脸图像的判别性的、稳定的低维特征表示更加容易了。像的判别性的、稳定的低维特征表示更加容易了。l 实验在实验在FERETFERET、ORLORL、ARAR、GTGT数据集上比较了提出的方法和其他数据集上比较了提出的方法和其他流行的子空间方法,表明我们的方法一致的优于其他方法。流行的子空间方法,表明我们的方法一致的优于其他方法。14谢谢观赏2019-5-9提纲l文章信息文章信息l背景介绍背景介绍本文解决的问题特征谱模型子空间分解l本文方法本文方法l实验结果实验
22、结果15谢谢观赏2019-5-9本文解决的问题l目标目标提取有判别力、稳定的特征用于分类l最常用的一类方法:最常用的一类方法:线性子空间的方法(linear subspace analysis)eg:LDAl存在的问题存在的问题Sw会出现奇异的情况Sw的特征值是从样本估计出来的,存在偏差和过拟合,推广能力差(poor generalization).argmaxTBoptTWWW S WWW S W16谢谢观赏2019-5-9相关方法l 现有的解决方法现有的解决方法去掉去掉Sw的一个子空间,使得的一个子空间,使得Sw不再奇异不再奇异。Nullspace LDA:去掉零空间,只在主空间中进行LD
展开阅读全文