基于Fisher准则改进线性判别回归分类的人脸识别

摘要

为了提高线性回归分类LRC（Linear Regression Classification）算法的鲁棒性，提出一种基于Fisher准则改进的线性判别回归分类算法。首先根据Fisher准则最大化类间重建误差与类内重建误差的比值，为LRC找到最优投影矩阵；然后利用最优投影矩阵将训练图像及测试图像投影到各个类的特征子空间；最后，计算出测试图像与各个训练图像之间的欧氏距离，并利用K-近邻分类器完成人脸的识别。在FERET和AR人脸数据库上的实验验证了本文算法的有效性。实验结果表明，相比其他回归分类算法，该算法取得了更好的识别效果。%Toimprovetherobustnessoflinearregressionclassification(LRC)algorithm,weproposealineardiscriminantregression classification algorithm which is improved based on Fisher criterion.First,it maximises the ratio of between-class reconstruction error and within-class reconstruction error according to Fisher criterion so as to find the optimal projection matrix for LRC.Then it makes use of the optimal projection matrix to project all training and testing images to subspace of every class.Finally,it computes Euclidean distances between the testing images and each training image and uses K-nearest neighbour classifier to finish face recognition.The effectiveness of the proposed algorithm is verified by experiment on FERET and AR face databases.Experimental results show that the proposed algorithm achieves better recognition effect compared with other regression classification approaches.