Face Recognition Using Nonlinear Regression

Face recognition algorithms mainly differ in how to represent the probe face image using the training data. As the state-of-the-art face recognition algorithm, linear regression computes a reconstruction matrix from the images of each subject and then approximates the probe face image using the reconstruction matrix. However, the performance of this linear algorithm is limited due to the nonlinear structure of the face images which is caused by variations in illumination, expression, pose and occlusion. To overcome the problem, in this paper we propose a kernel-based nonlinear regression algorithm for effective face recognition. Because of the high (even infinite) dimensionality of the nonlinear transformation function, it is infeasible to directly calculate the corresponding reconstruction matrix and therefore is unable to explicitly approximate the probe image. With the help of kernel trick, we tackle this difficulty by embedding the nonlinear regression in the stage of computing the distance between the probe image and the approximated probe image. The proposed nonlinear regression classification algorithm is evaluated on several popular standard databases under a number of classical evaluation protocols that have been reported in the face recognition literature. A comparative study with linear regression classification approach and several other algorithms shows the superiority of the proposed approach.