Learning from Face recognition under occlusion

Occluded face recognition is a key problem to be solved in a reality-oriented face recognition system. From the point of view of subspace regression, this paper reviews the general methods of mainstream linear regression classifiers to deal with occlusion problems: collaborative representation, dictionary representation of occlusion, and learning and compression techniques of occlusion dictionaries. The existing iterative weight error coding methods are reviewed from two aspects of noise suppression and occlusion detection. In this paper, two ideas to solve the occlusion problem in a modern face recognition system are proposed: the restoration of the occlusion region and the location of occlusion based on attention. The application systems for these two ideas are given respectively. This paper emphasizes the importance of feature extraction in solving the problem of occluded face recognition. Finally, it points out the challenges that occluded face recognition brings to computer vision; the main problems existing in existing methods in optimization algorithms and feature extraction, and the key problems to be considered in using convolution neural networks to deal with occlusion problems in the future. It is introduced that solving the problem of face recognition under occlusion brings many benefits to life.