Graph Regularized Weighted Low-Rank Representation for Image Clustering

Low-rank representation have been widely used in computer vision processing. In this paper, we propose a new models of graph regularized low-rank representation that introduce the regularization term of graph which can describe the similarity between samples. In order to further enhance the low rank of the matrix, we propose a graph regularization weighted low-rank representation model which introduce weighted nuclear norm. The purpose of this paper is to study the influence of the low-rank representation of the regularization term of graph on the image clustering effect. The dataset used in this experiment is USPS (United States Postal Service) handwritten digits data set. The clustering algorithm is a k-means algorithm. Two types of experiments were done for each model: clustering of two classes of numbers and clustering of three classes of numbers. Similarly, this paper also compares the clustering effects of these two models. Finally, the experimental results show that the graph regularized can enhance the clustering effect of the image obviously.