摘要： 为解决人脸识别中的维数灾难问题，提出一种基于自适应核边际费希尔分析的人脸识别算法。在考虑图像流形结构的基础上给出与图像数据相关的自适应核函数，采用核边际费希尔分析对高维人脸图像进行非线性降维，利用最小二乘支持向量机在降维后的低维特征空间中进行分类识别。实验结果表明，该算法的识别性能优于其他常用的人脸识别算法。

关键词: 人脸识别, 核边际费希尔分析, 流形学习, 最小二乘支持向量机, 降维, 核理论

Abstract: To effectively cope with the curse of dimensionality problem of face recognition, a face recognition algorithm based on adaptive Kernel Marginal Fisher Analysis(KMFA) is proposed in this paper. The adaptive kernel function which closely relates to image data is first introduced by considering the image manifold structure, then the high-dimensional face images are nonlinearly reduced to the lower-dimensional feature space with KMFA. Least Squares Support Vector Machine(LS-SVM) is used to classify and recognize different face image herein. Experimental results show that the proposed algorithm performs much better than other traditional face recognition algorithms.

Key words: face recognition, Kernel Marginal Fisher Analysis(KMFA), manifold learning, Least Squares Support Vector Machine(LS-SVM), dimensional reduction, kernel theory

中图分类号: 

• TP181