Tensor Linear Regression and Its Application to Color Face Recognition

Linear regression has achieved the promising preliminary results for face classification. But, most existing methods are incapable of tackling color images classification. The major reason is that they need to transform each color image to a vector or matrix, leading to the loss of multidimensional structure information embedded in color images. To address this problem, we study the tensor linear regression problem, and develop a novel tensor low-rank method, which utilizes tensor-Singular Value Decomposition (t-SVD) based tensor nuclear norm to emphasize the spatial structure embedded in color images. Applying it to color face classification, extensive experiments on three datasets demonstrate that our method is superior to several state-of-the-art methods.