Abstract: Since the dimension of face features which is presented by Gabor wavelet is too high, there has large computation and bad real-time problems if using the feature by Gabor wavelet transform for recognition directly. A new face recognition algorithm based on Gabor wavelet transform and modular PCA（Principal Component Analysis） is proposed. Face image feature is acquired by Gabor wavelet transforming face image. Its dimension is reduced and eigenvectors are extracted by the method of modular PCA. Nearest neighbor classifier is adopted to sort and distinguish. Experimental results on ORL and NUST603 indicate that the performance of proposed method is superior to other methods, such as PCA, modular PCA and combination of Gabor wavelet transform and PCA.

Key words: face recognition, Gabor wavelet, modular Principal Component Analysis（PCA）, feature extraction

摘要： 由于Gabor小波描述的人脸特征维数太高，直接将Gabor小波提取的特征进行识别时出现计算量大、实时性差的问题，提出了基于Gabor小波变换与分块主分量分析的人脸识别新算法。首先对人脸图像进行Gabor小波变换得到人脸图像特征，然后用分块主分量分析方法对其进行降维、提取特征向量，最后用最近邻分类器分类识别。在ORL和NUST603人脸库上进行实验，结果表明，该方法的识别率优于传统PCA、分块PCA、Gabor小波变换与PCA结合的方法。

关键词: 人脸识别, Gabor小波, 分块主分量分析（PCA）, 特征提取