Enhancing Kernel Maximum Margin Projection for Face Recognition

To efficiently deal with the face recognition problem, a novel face recognition algorithm based on enhancing kernel maximum margin projection(MMP) is proposed in this paper. The main contributions of this work are as follows. First, the nonlinear extension of MMP through kernel trick is adopted to capture the nonlinear structure of face images. Second, the kernel deformation technique is proposed to increase the discriminating capability of original input kernel function. Third, the feature vector selection approach is applied to improve computational efficiency of kernel MMP. Finally, the multiplicative update rule is employed to enhance training speed of SVM classifier for face recognition. Experimental results on face recognition demonstrate the effectiveness and efficiency of the proposed algorithm.