Orthogonal Procrustes Problem Based Regression with Application to Face Recognition with Pose Variations

Recently, sparse representation and collaborative representation based classifiers for face recognition have been proposed and achieved great attention. However, the two linear regression analysis based methods are sensitive to pose variations in the face images. In this paper, we combine the orthogonal Procrustes problem (OPP) with the regression model, and propose a novel method called orthogonal Procrustes problem based regression (OPPR) for face recognition with pose variations. An orthogonal matrix is introduced as an optimal linear transformation to correct the pose of test image to that of training images as far as possible. According to where the matrix is multiplied to deal with pose variations in vertical or horizontal direction, we propose the left or the right side OPP based regression, respectively. What’s more, we further fuse the two models and propose a bilateral OPP based regression. The proposed model is solved via the efficient alternating iterative algorithm and experimental results on public face databases verify the effectiveness of our proposed models for handling pose variations.