基于局部本征谱的人脸识别

Local Eigenspectra-Based Face Recognition

设计了一种基于局部本征谱的人脸识别方案。预处理阶段,首先将一幅脸像按不同方位划分为大小相同的数个子块,针对各子块进行能量归一化和傅里叶变换,以消除部分光照影响并估算子块的频谱。在此基础上,对训练脸像中编号相同的子块进行主元分析,提取脸像的局部本征谱,采用最近邻判决准则进行分类识别。对ORL人脸数据库的实验结果表明本设计方案是有效的。

This paper addresses a local eigenspectra-based scheme for face recognition, wherein each face is partitioned into a suitable number of blocks, followed by energy normalizing to reduce the brightness variation effect and by the Fourier transform to estimate the spectra of each block. Features called eigenspectra are obtained by the principal component analysis（PCA） on the same serial number blocks, and then classified by the nearest neighbor（NN） rule. Experiments taken on the Olivetti Research Laboratory（ORL） face database show the feasibility of the addressed method for face recognition.