Nuclear norm based adapted occlusion dictionary learning for face recognition with occlusion and illumination changes

Abstract

In this paper, we propose a robust framework named Nuclear Norm based adapted Occlusion Dictionary Learning (NNAODL) for face recognition with illumination changes and occlusions. Specifically, we first introduce a nuclear norm based error model to characterize the occlusion and corrupted region in the query image. Secondly, we innovatively integrate the error image with training samples to construct the dictionary, thus can both accurately reconstruct the corrupted region and non-corruption region in query images. Moreover, we use two-dimensional structure for representation and adapted sample weights to preserve more structural information. Above advantages are integrated by a unified objective function, and an effective algorithm is proposed to solve our model. Compared with existing sparse representation methods, our model can better represent the noisy samples and reduce the influence of occlusion and pixel errors. Experiments on multiple public datasets show that the NNAODL model can achieve better results than classical methods under occlusion and illumination changes.