Multi-Graph Constraint Matrix Factorization for Multi-view Image Clustering

Multi-view clustering aims to obtain the true cluster with a set of multiple views of data. NMF-based approach, a type of widely used methods, can learn unified common agreement from multiple sources of information, and gives a relatively good performance. However, these methods usually ignore the complementary information in each view, which is important to learn the structure of data. In this paper, each view was treated as a subset feature of data, and a novel multi-graph constraint semi-nonnegative matrix factorization was proposed to learn the overall representation graph of data, and the spectral clustering was used to cluster data finally. Experiments on classic images datasets have shown that our method can improve the multi-view clustering performance clearly compared to other classic methods.