摘要
正交鉴别保局投影算法是一种有效的特征提取方法,但是将其应用在人脸识别中将遇到小样本问题,并且算法只是一种线性的特征提取方法,因此提出一种核正交鉴别保局投影算法.实现这一算法的关键是高维特征空间中总体散布矩阵的非零空间的计算,对此根据eigenfaces中将高阶矩阵计算转化成低阶矩阵计算的思想及核函数技术,将高维特征空间中总体散布矩阵的非零空间的计算仍归结为标准的特征值分解问题,并且所提出的算法能够有效地解决小样本问题.在人脸库上的实验结果验证了所提出的算法是可行的和有效的.
Orthogonal discriminant locality preserving projections is an effective feature extraction,but it may encounter the small size samples problem when it is applied in face recognition task.In addition,it is only a linear feature extraction technique.A kernel orthogonal discriminant locality preserving projections is proposed.The key is to how to compute the nonzero space of the total scatter matrix in the higher dimensional feature space.As to this problem,the kernel function technique and the eigenfaces method that transforms the computation of the high order matrix into the computation of the low order matrix are used,and then the actual computation of the nonzero space of the total scatter matrix in the higher dimensional feature space is reduced to a standard eignenvalue problem.In addition,the proposed algorithm can effectively overcome small size samples problem.The numerical experiments on facial databases show that the proposed method is effective and feasible.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2010年第4期979-982,965,共5页
Acta Electronica Sinica
基金
中国博士后基金(No.20060400809)
黑龙江省青年科技基金(No.QC06C022)
关键词
正交鉴别保局投影
特征提取
人脸识别
总体散布矩阵
核函数
orthogonal discriminant locality preserving projections
feature extraction
face recognition
total scatter matrix
kernel function