基于核的最近邻特征重心分类器及人脸识别应用

Kernel-based nearest neighbor feature centroid classifiers for face recognition

本文提出核最近特征线和特征面分类器,可直接对高维人脸图像进行识别。为解决计算量大和可能失效的问题,提出(核)最近特征重心和(核)最近邻特征两种解决方法,前者降低了计算特征线和面距离的复杂度,后者减少了特征线和面的数目,两种方法均避免了可能失效的问题。将二者结合得到的(核)最近邻特征重心分类器,在获得相近识别率的条件下,使计算复杂度降到了最小。所得方法无需预先抽取人脸图像特征,因此避免了在较多样本数时,特征抽取存在计算量大的问题。基于ORL人脸数据库的实验验证了本文方法的有效性。

The kernel-based nearest feature line and plane classifiers are proposed in this paper. To overcome the problems of large computational cost and possible failure, the （kernel） nearest feature centroid and （kernel） nearest neighbor feature classifiers are presented. The former reduces the cost in computing feature line and plane distance, and the later reduces the number of feature line and plane. Both methods avoid the possible failure. The （kernel） nearest neighbor feature centroid classifiers are proposed to further reduce the computational cost. The proposed methods need not extract the facial features, which therefore can directly classify original face images. Experimental results on ORL face database demonstrate the feasibility of the proposed methods.