人脸识别算法研究—基于图像多路正交投影和最小类距离分类器  

Face Recognition Algorithm Based on Image Multi-Orthogonal Projection and Least Class Distance Classifier

作　　者：毛家发[1] 马燕[2] 周玉林[1] 熊鹏荣[1] 

机构地区：[1]上饶师范学院计算机系,上饶334001 [2]上海师范大学数理信息学院,上海200234

出　　处：《计算机工程与应用》2005年第27期77-79,共3页Computer Engineering and Applications

基　　金：上饶师院科研基金资助课题

摘　　要：论文提出了基于图像多路正交投影和最小距离分类器的人脸识别方法。该方法与Fisherfaces方法相比,是基于图像矩阵的,故它的突出优点是极大地提高了特征提取的速度;与最近邻分类法相比,论文提出的最小类距离分类法,在识别性能上有明显的改善。最后,在ORL标准人脸库上的实验结果表明,其特征提取的速度是Fisherfaces特征提取速度的14倍;在相同的特征维数下,识别性能与最近邻法相比也得到较大的提高。This paper proposes an image multi-orthogonal projection method for face recognition based on least class distance classifier.Compared with Fisherfaces method,it has a desirable property.It is directly based on image matrices, so the time of feature extraction will be saved.And compared with the nearest classifier method,the proposed method improves the accuracy of recognition.Finally,the proposed method is tested on ORL face databases.The experimental results indicate that the speed of feature extraction by the proposed method is nearly 14 times faster than Fisherfaces, and the ability of discriminant method is powerful than the nearest classifier.

关 键 词：图像多路正交投影 最小类距离分类器 Fisherfaces 方法 图像特征提取 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]