摘要
在人脸识别应用中,当每个人有多个训练样本(MSPP)时,Fisher线性判别分析(FLDA)方法可以很好地用于特征提取。然而,当每个人只有一个训练样本(SSPP)时,因为类内散布矩阵为零矩阵,所以FLDA方法将不能使用。为了解决该问题,提出了一种比较新颖的方法来估计类内散布矩阵,借助于奇异值分解(SVD)方法,先将人脸图像分解成两部分,然后分别估计出类内散布矩阵及类间散布矩阵,使FLDA方法能够得到有效的应用。在ORL及Yale上的实验表明了提出的方法比现有的许多方法取得了更好的识别效果。
Usually, Fishier Linear Discriminative Analysis （FLDA） can be effective in face recognition when each person has multiple samples （MMSP）. However,it will not be used when each person has only one training sample （SSPP） because the intra-class metric is zero. To address this problem ,a novel method to estimate the intra-class scatter metric is proposed. By using the Singular Value Decomposition （SVD） ,firstly, face image is decomposed into two parts,and then they are used to estimate intra-class and inter-class scatter metrics,which making the traditional FLDA can be applied to SSPP task. Experiments on the ORL and Yale face database show that the proposed method can achieve better recognition accuracy than many common solutions to the SSPP problem.
出处
《电视技术》
北大核心
2013年第15期181-184,共4页
Video Engineering
关键词
单训练样本每人
奇异值分解
FISHER线性判别分析
face recognition
single training sample per person
singular value decomposition
fishier linear discriminative analysis
