摘要： 提出一种简单且识别率高的分块分色的人脸图像特征抽取方法。该方法将人脸图像按行均等分块，在HSV颜色空间抽取每个子块图像的每维颜色空间模式值的均值和方差特征，用改进马氏距离的最近邻法对人脸图像进行分类，并采用留一法进行交叉验证。通过实验发现，对ORL、faces94、faces95这3个常用的人脸图像库，都能取得超过99.5%的识别正确率。

关键词: 分块, 分色, 改进的马氏距离, 最近邻法, 留一法

Abstract: Feature extraction method of face image, which boasts simplicity and high recognition rate, is proposed. Partitioning face image by row equally, this method uses mean value of dimensional color space pattern values extracted from each sub-block image in HSV as well as variance, classifying face images with nearest neighbor method in improved Mahalanobis distance method and cross-validated through leave-one-out method. Through test, it turns out that this method can get more than 99.5% identification accuracy in such three common face database as ORL, faces94, faces95.

Key words: partition, color separation, improved Mahalanobis distance, nearest neighbor method, leave-one-out method

中图分类号: 

• TP391.41