Nonlinear Metric Learning with Deep Convolutional Neural Network for Face Verification

Face verification is a very challenge problem, due to large variations in expression, background, pose, and occlusion. It involves two crucial problems, one is face representation and the other is the similarity computation of face vectors. Addressing the two problem, this paper proposes a method for simultaneously learning features and a corresponding similarity metric for a real world face verification, which apply novel regularization to learn a nonlinear metric learning with deep convolution neural network. Experimental results on the widely used LFW dataset are presented to show the effectiveness of the proposed method.