基于Fisher的线性判别回归分类算法

Linear Discriminant Regression Classification Algorithm by Fisher Criterion

为了提高线性回归分类(LRC)算法的鲁棒性,提出了一种基于Fisher准则的线性判别回归分类算法。利用Fisher准则将类间与类内重建误差的比值最大化,找到线性回归分类的最优投影矩阵;再将训练图像及测试图像投影到各类的特征子空间;求得各训练图像与测试图像间的欧氏距离,最后用K-近邻分类器完成人脸识别。在AR人脸数据库上的实验结果表明,相比其他回归分类算法,算法取得了更好的识别效果。

To improve the robustness of the linear regression classification （LRC） algorithm, a linear discrimi？nant regression classification algorithm based on Fisher criterion is proposed. The ratio of the between-class re？construction error over the within-class reconstruction error is maximized by Fisher criterion so as to find an opti？mal projection matrix for the LRC. Then, all testing and training images are projected to each subspace by the op？timal projection matrix and Euclidean distances between testing image and all training images are computed. Fi？nally, K-nearest neighbor classifier is used to finish face recognition . Experimental results on AR face databases show that proposed method has better recognition effects than several other regression classification approaches.