Abstract: Three kinds of features including Gabor amplitude, Gabor phase and combination of amplitude and phase features are extracted, PCA is used to reduce the dimensions, and the nearest neighbor classification rule is used for recognition. Recognition results of the three methods are analyzed and compared on six databases including ORL, Yale, Indian, YaleBE, PIE and the FERET. The results show that Gabor wavelet phase features are robust to light variation, which outperforms other methods on YaleBE and Yale datasets. Gabor wavelet amplitude and phase combined feature is more robust to expression and time changes, which gets higher recognition rate on the fb, dup1, dup2 probe sets of FERET dataset.

Key words: face recognition, Gabor wavelet, Gabor amplitude and phase features, Principal Component Analysis（PCA） dimension reduction, nearest neighbor classification

摘要： 对采用Gabor幅值、Gabor相位以及Gabor幅值加相位结合这三种方法，在同等条件基础上对所提取的特征进行分块，用PCA降维，采用最近邻分类规则进行识别并比较结果。在ORL、Yale、Indian、YaleBE、PIE和FERET六个数据库进行比较研究的结果表明，Gabor小波相位特征对光照有较高的鲁棒性，在光照变化明显的Yale和YaleBE数据库识别效果最好，而Gabor小波幅值加相位特征具有表情和时间变化的鲁棒性，在FERET的fb、dup1、dup2测试集上获得了较高的识别率。

关键词: 人脸识别, Gabor小波, Gabor幅值和相位特征, 主成分分析（PCA）降维, 最近邻分类