针对传统的人脸识别方法对人脸图像的曝光量、表情比较敏感,并且具有较大类内离散度的缺点,该文提出一种基于Gabor小波以及加权互协方差运算的人脸识别算法.该算法首先对人脸图像提取Gabor特征,然后使用加权的互协方差矩阵对经过处理的特征图像进行降维及特征提取;最后使用最近邻分类器进行分类.在ORL数据库和AR数据库上的实验表明,该方法的降维和识别性能优于传统2DPCA及其改进算法,能兼顾维度简约性和准确性,有效地提高了识别性能.%The traditional face recognition is sensitive to light condition as well as facial expression, and has a shortcoming of high intra-group dispersion, a novel method is proposed to overcome these defects by combining Gabor wavelet and a weighted computation based on the cross-covariance. Firstly, Gabor features are extracted from the face image. Then, a weighted cross-covariance matrix is used for dimension reduction and feature extraction. Finally, the nearest neighbor classifier is performed for classification. Experimental results on the ORL face database and the AR face database show that the recognition performance of the proposed method is superior over the 2DPCA and its improved algorithm. It also reduces the dimensionality of feature and improves the recognition performance effectively.
展开▼
