Multi-View Clustering With Consistent Local Structure-Guided Graph Fusion

Abstract

With the development of camera and sensor technologies, multi-view data are ubiquitous and require more technologies to process them. Multi-view clustering with graph fusion has recently attracted considerable attention as multiple graphs defined by views can provide more comprehensive information for clustering. Different from previous methods that rarely consider the locality of the fused graph, in this paper, we propose an $\ell _{0}$-norm constrained graph fusion model with the ability to preserve the consistent local structure of the fused graph, as well as the view weights which are obtained adaptively. Also, to solve the proposed model, we design an efficient algorithm with a closed-form solution for each variable, together with the analysis of the convergence. Experimental results indicate that the learned consistent local structure can refine and guide the graph fusion to achieve a better graph, and our method outperforms the state-of-the-art graph fusion methods.