Deep learning in face recognition across variations in pose and illumination

Even though face recognition in frontal view and normal lighting conditions works very well, the performance drops sharply in extreme conditions. Recently there is plenty of work dealing with pose and illumination problems, respectively. However both the lighting and pose variations always happen simultaneously in general conditions, and consequently we propose an end-to-end face recognition algorithm to deal with two variations at the same time based on convolutional neural networks. In order to achieve better performance, we extract discriminative nonlinear features that are invariant to pose and illumination. We propose to use the 1?×?1 convolutional kernels to extract the local features. Furthermore a parallel multi-stream convolutional neural network is developed to extract multi-hierarchy features which are more efficient than single-scale features. In the experiments we obtain the average face recognition rate of 96.9% on MultiPIE dataset. Even for profile position, the average recognition rate is also around 98.5% in different lighting conditions, which improves the state-of-the-art face recognition across poses and illumination by 7.5%.