对称DLDA及其在人脸识别中的应用

针对直接线性鉴别分析(DLDA)没有有效利用人脸对称性特征,及其在人脸识别中训练样本不足的问题,依据人脸较为明显的镜像对称性,结合该特性在直接线性鉴别分析的基础上提出对称直接线性鉴别分析方法。采用镜像变换得到奇对称样本和偶对称样本,再分别提取各奇偶对称样本特征分量,最后采用最小欧氏距离进行分类。通过在ORL和YALE人脸数据库上的实验证明,该算法不仅有效利用了镜像样本,扩大了训练样本容量;而且取得了比直接线性鉴别分析更好的识别性能。

Symmetrical DLDA and Its Application to Face Recognition

Aiming at the problem that Direct Linear Discriminant Analysis (DLDA) does not effective use the feature of face symmetry, and face recognition are generally lack of training samples. According to the obvious feature of face symmetry, this paper proposes a face recognition method based on DLDA, the Symmetrical DLDA. The odd symmetry samples and the even symmetry samples are received by mirror transforming. After that, odd/even symmetrical principal components are respectively extracted. Lastly, Minimum Euclidean Distance is employed to classify the extracted features .The method is evaluated on the ORL and YALE face image database. Experimental results show the proposed method not only effective use of mirror samples, enlarges the number of training samples, but also achieves better performance than DLDA.