摘要
在局部判别嵌入的基础上提出了一种有效的非线性子空间学习方法:类别多核局部判别嵌入.首先针对给定数据的类别信息,定义基于每一个类别的局部核函数,形成多核,接着将不同的局部核函数进行线性组合作为最终的核函数引入到局部判别嵌入算法中,得到类别多核局部判别嵌入算法,在核空间内提取图像高阶非线性信息.在ORL和Yale库上的人脸识别表明该方法是有效的.
Based on local discriminant embedding, an efficient nonlinear subspace learning method, Label Multiple Kernel Local Discriminant Embedding （LMKLDE） , is developed. Firstly, according to the label information of given data set, local kernel function is defined and multiple kernel is gained. Then, different local kernel functions are merged by linear combination to form final kernel function. Finally, LMKLDE is developed by introducing label multiple kernel to LDE in order to deal with datasets of highly nonlinear structure. Experiments on ORL and Yale face database demonstrate the effectiveness of the proposed method.
出处
《郑州大学学报（工学版）》
CAS
北大核心
2012年第3期125-128,共4页
Journal of Zhengzhou University（Engineering Science）
关键词
人脸识别
子空间学习
类别多核
降维
face recognition
subspace learning
label multiple kernel
dimensionality reduction
