Subspace learning based on label release and low-rank representation for small sample face recognition

Subspace learning is often used to solve face recognition problems, and has achieved good results in some scenes. However, the current methods based on subspace learning still have the following problems: As a kind of important information, label information is often ignored in the process of model building. Besides, many methods lack the theoretical basis for enlarging the difference between classes. To solve these problems, this paper proposes a subspace learning method based on Label Release and Low-rank Representation (LRLR), and applies this method to small sample face recognition. In LRLR, on the one hand, we use label information to build a label release model, and embed this model in the process of subspace learning, so that the learned mapping matrix can map samples to the new subspace to achieve the purpose of increasing the difference between classes. On the other hand, the nuclear norm and sparse norm are used to protect the intrinsic structure of the data. Experimental results show that the proposed LRLR is effective in face recognition.