基于流形的表情分解算法.pdfVIP

维普资讯 ComputerEngineeringandApplications计算机工程与应用 2008，44(23) 203 基于流形的表情分解算法 朱明旱 2，罗大庸 1 ZHU Ming—ban1,2LUO Da-yong 1．中南大学 信息科学与工程学院，长沙 410083 2．湖南文理学院 电气系，湖南 常德 415000 1．CollegeofInformation Science and Engineering，CentralSouth University，Changsha410083，China 2．DepartmentofElectronics，Hunan University ofArtsand Science，Changde，Hunan415000，China E—mail：zhumh 123@163．con 一 ZHU M ing—han，LUO Da-yong．Expression decomposition algorithm based on appearancemanifolds．ComputerEngineering andApplications，2008，44(23)：203-205． Abstract： In this paper，expression decomposition algorithm based Oilappearance manifolds is pooposed．Firstly，face images are mappedintolow—dimensionalmanifoldsubspaceusingLocalityPreservingProjections(LPP)．Secondly,theselow—dimensionalrepre～ sentationsare decomposed into person subspace and facialexpression subspace．Finally，face and facialexpression recognition are realized．The problem ofimagefeatureextraction in higher-ordersingularvaluedecomposition issolved by usingappearance man— ifoldslearningmethod．ByapplyingHigher-OrderSingularValueDecomposition(HOSVD)toappearancemanifolds，personpattern hardly influence facialexpression recognition．The advantagesofhigher-ordersingularvalue decomposition and manifoldslearning arefused effectively．Thealgorithm istested on JAFFE and CMP-AMP face database．The experimentalresultshowsthatitis powerfuland effectiveforface and facialrecognition． Keywords：appearancemanifolds；LocalityPreservingProjections(LPP)；Higher-OrderSingularValueDecomposition(HOSVD)； face recongition；expression recognition 摘 要：提 出了基于流形的表情分解算法。首先，运用保局投影将图像投影到低维的表情流形子空间，再在流形子空间里对它们进 行高阶奇异值分解，最后在个人子空间和表情子空间里完成人脸和表情识别。该算法用流形学习解决了高阶奇异值分解中的图像 特征值提取问题，用高阶奇异值分解解决了流形表情识别中个人模式影响表情识别的问题。是一种流形学习与高阶奇异值分解优 势互补的算法。在CMU—AMP和JAFFE人脸库上的实验表明，该算法对人脸和表情识别都十分有效。 关键词：表情流形；保局投影；高阶奇异值分解；人脸识别；表情识别 DOI：10．3778／j．issn．1002—8331．2008．23．061 文章编号：1002—8331(2008)23—0203—03 文献标识码