采用Gabor变换和双方向LDA进行人脸识别

Face Recognition Based on Gabor Transform and Bidirectional LDA

将Gabor变换和双方向LDA(BDLDA)算法相结合,提出了一种可以解决小样本问题的人脸识别新算法.该算法把人脸图像经过Gabor小波变换后得到的每个输出图像都看成是独立的样本,使得每类人脸样本的样本数成倍增加.然后采用BDLDA算法来提取人脸特征,并专门设计了针对人脸特征矩阵的最近邻分类器和最小距离分类器来进行分类判决.在ORL人脸库和FERET人脸库中的实验结果表明,当每类的训练样本数较少时,该算法能大幅度提高人脸识别率,甚至当每类训练样本数仅为1时,也能得到较高的性能.

A novel face recognition algorithm can solve the small sample size （SSS） problem by integrating Gabor transform and bidirectional LDA （BDLDA）. The sample size of each subject is multiplied by using every output image after taking Gabor wavelet transform as an independent sample. The BDLDA method is adopted for face feature extraction. Then, special nearest neighbor classifier and minimum distance classifier based on face feature matrix are designed for classification, respectively. The experimental results on ORL face database and FERET face database show that the proposed method can increase the face recognition rate greatly when the training sample size is small, and can get a good performance even when the training Sample size of each subject is only 1.