基于图像分块的改进Fisher人脸识别算法  

Improved Fisher algorithm for face recognition based on image segmentation

作　　者：梁淑芬[1,2,3] 甘俊英[1,2,3] 

机构地区：[1]五邑大学信息学院,广东江门529020 [2]北京大学视觉与听觉信息处理国家重点实验室,北京100871 [3]浙江大学CAD&CG国家重点实验室,杭州310027

出　　处：《计算机工程与应用》2009年第19期165-167,共3页Computer Engineering and Applications

基　　金：广东省自然科学基金(No.032356;No.07010869);北京大学视觉与听觉信息处理国家重点实验室开放课题基金项目(No.0505);浙江大学CAD&CG国家重点实验室开放课题基金项目(No.A0703);江门市科技攻关项目(No.江财企[2004]59号)~~

摘　　要：二维方法用于图像矩阵特征提取,虽然速度快,但影响了分类速度。针对二维线性鉴别分析(Two-Dimensional Linear Discriminant Analysis,2DLDA)的特点,研究了一种基于图像分块的改进Fisher人脸识别算法,该算法首先对人脸图像进行压缩降维处理,得到相应的特征矩阵,然后利用改进Fisher算法对特征矩阵进行类间离散度矩阵和类内离散度矩阵的计算,该算法充分考虑了类别信息,避免了传统Fisher算法造成的小样本问题,有效提高了分类速度。基于ORL(Olivetti Research Laboratory)与Yale人脸数据库的实验结果证明了该算法的有效性。The features of original image matrixes can be computed directly and rapidly by way of two-dimensional methods,but the classification speed is affected by many features.Combined with the traits of Two-Dimensional Linear Discriminant Analysis （2DLDA）,a method of improved Fisher algorithm for face recognition is presented based on image segmentation.Firstly, compression is done to face image respectively,and the corresponding feature matrixes are obtained.Secondly,The improved Fisher algorithm works with matrix representation.Compared with the traditional Fisher method,small samples problem are avoided by the improved Fisher method,in which category information is considered and classification speed is improved.Experimental results on ORL（Olivetti Research Laboratory） and Yale face database show that the propsed method is valid in face recognition.

关 键 词：人脸识别 二维线性鉴别分析 改进Fisher算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]