摘要
因为Gabor特征的维数很高并且存在大量信息冗余,所以很有必要研究合适的降维算法以降低Gabor特征的维数.为了解决这个问题,提出了最优Gabor尺度和方向的选择算法.在这个算法中,把所有的样本和每一个Gabor核进行卷积,并对所有的卷积结果分别计算类内距离和类间距离.最后,通过计算类间距离和类内距离的比值选择比值最大的Gabor核就是对应的最优Gabor核.为了验证本文算法的有效性,分别在YALE、AR、FERET人脸数据库上进行实验,结果表明较大尺度和某些方向构成的Gabor核对应的特征具有较好的鉴别力.
Because the fact that Gabor features are redundant and too high-dimensional, appropriate feature dimension reduction appears to he much more necessary. To address this problem, a novel optimal selection method of Gabor kernels＇ scales and orientation was proposed. In this method, all training samples were convolved with each Gabor kernel. Within-class distance and between-class distance calculation were performed on these convolution results, respectively. At last, the optimal Gabor kernel was selected based on the ratio of the within-class distance and the between-class distance, in which Gabor kernel corresponding to the largest ratio is the optimal one. To prove the advantages of the proposed method, extensive experiments were conducted on popular face databases such as YALE, AR, FERET. The experiment results show that the proposed method is effective and the features in the larger scales as well as the features in 135°, 180° and 225°orientations have more discriminative power.
基金
Supported by the National Science Foundation of China(61102040)
关键词
人脸识别
Gabor核
降维
face recognition
Gabor kernel
dimension reduction
