一种自适应加权的基于Gabor滤波器的人脸识别方法

摘要

In order to improve the performance of the recognition based on Gabor filter, an adaptive feature weight selection method is proposed in this paper. First, the even symmetry samples are obtained by mirror transforming, which constitutes detection image set. Secondly, the dimension of each face image is reduced by DCT. A feature images is obtained by Gabor filters. Then the contribution of each feature could be adaptive-ly computed, and be weighted to identify features. Finally, the nearest neighbor classifier is used for identification. The experiment result on ORL and Yale face databases shows this method is effective.%为了改善Gabor滤波器的识别性能,提出了一种自适应的基于Gabor滤波器的特征权重选择的人脸识别方法.首先将训练样本进行镜像变换,由镜像偶对称图像构成探测图像集,然后把每幅人脸图像采用离散余弦变化进行降维,经过Gabor小波变换提取人脸特征,再自适应地计算出不同特征对识别的不同贡献并加权到鉴别特征中,最后根据最近邻分类器分类.基于ORL和Yale人脸库上的实验结果验证了改进算法的有效性.