View-specific anchors coupled tensorial bipartite graph learning for incomplete multi-view clustering

Abstract

The incomplete multi-view clustering (IMVC) aims to explore the consensus and complementary information embedded in incomplete multi-view data, so as to accurately aggregate all samples into different clusters. Existing IMVC methods still have the following issues that need to be further solved: how to efficiently (1) alleviate the computational complexity; (2) capture the complementary information from views; (3) uncover the high-order correlations underlying incomplete multi-view data. To address these problems, a new method, called view-specific anchors coupled tensorial bipartite graph learning for incomplete multi-view clustering (VA-TBGIMC), is proposed in this paper. Specifically, view-specific anchors are learned from non-missing samples to represent the distribution of all samples including missing ones. With the help of high-quality anchors, the bipartite graph between all samples and anchors is constructed, which can preserve the complementary information. Meanwhile, these bipartite graphs are stacked into a tensor which is imposed with a low-rank constraint to adequately capture the inter-view and inter-sample correlations from incomplete multi-view data. Further, the augmented graph of bipartite graph is constrained by Laplacian rank to learn a consensus cluster indicator. A large number of experiments validate the superiority and efficiency of the proposed method in terms of clustering performance and time consumption.