关键词: 非负矩阵分解, 自表示, 图学习, 判别信息, 人脸识别

Abstract: Abstract: Partial representation-based non-negative matrix factorization technology and its variants have been widely used in the fields of information retrieval, computer vision and pattern recognition. The most representative one is graph regularized non-negative matrix factorization algorithm, which makes full use of the assumption that high-dimensional data is usually located in a low-dimensional manifold space to construct Laplacian matrix. However, the disadvantage of the regular non-negative matrix factorization based on the graph is that the constructed Laplacian matrix is calculated in advance, and it is not iterated during the multiplicative updating process. In order to solve this problem, this paper combines the self-representation method in subspace learning to generate the representation coefficients, and further calculates the similarity matrix to obtain the Laplacian matrix, and iterates the Laplacian matrix during the update process. In addition, the class indicator matrix is constructed using the label information of the training set, and two different regular terms are introduced to reconstruct the class indicator matrix. The method proposed in this paper is called graph learning regularized discriminative non-negative matrix factorization (GLDNMF). We also give the corresponding multiplicative updating solutions and the convergence proof for the optimization framework. Face recognition experiments on two standard datasets show the effectiveness of the proposed method.

Key words: Non-negative Matrix Factorization, Self-representation, Graph Learning, Discriminative Information, Face Recognition