Abstract: This paper presents a new feature extraction method for face recognition using 2D Circularly Symmetrical Gabor Transform（2DCSGT） and Independent Component Analysis（ICA）.A circularly symmetrical Gabor feature vector is derived from a set of downsampled circularly symmetrical Gabor wavelet representations of face images.The dimension of the circularly symmetrical Gabor feature vector is reduced by means of Principal Component Analysis（PCA）.Independent Circularly Symmetrical Gabor Features（ICSGF） are defined based on Independent Component Analysis.To show the validity of the proposed method，it is applied to face recognition on the ORL，YALE and FERET databases.In particular，the ICSGF method achieves 99.5% correct face recognition accuracy for ORL database，93.33% accuracy for Yale database and 97.14% accuracy for FERET database.Experimental results show that the algorithm is feasible and effective for face recognition.

Key words: face recognition, Circularly Symmetrical Gabor Transform（CSGT）, Principal Component Analysis（PCA）, Independent Component Analysis（ICA）, nearest neighbor classifier

摘要： 提出了一种基于2D环形对称Gabor变换（2DCSGT）和独立成分分析（ICA）的特征提取方法，并用于人脸识别中。对人脸图像做5尺度2D环形对称Gabor变换;对变换后的图像下采样并采用主成分分析（PCA）进行降维;采用ICA获得人脸图像的独立环形对称Gabor特征（ICSGF）;采用最近邻分类法分类。在ORL、YALE和FERET人脸数据库上，基于ICSGF的人脸识别算法的正确识别率分别达到99.5%、93.33%和97.14%。实验结果表明，ICSGF方法有效可行。

关键词: 人脸识别, 环形对称Gabor变换（CSGT）, 主成分分析（PCA）, 独立成分分析（ICA）, 最近邻分类器