Local structure learning for incomplete multi-view clustering

Incomplete multi-view clustering, which aims to divide different groups into incomplete views produced by various sensors, has attracted research attention. In this article, we propose a local structure learning for incomplete multi-view clustering (LS-IMC) algorithm. The algorithm jointly learns a consensus of incomplete views and a clustering result. Specifically, by fusing consistent representation and local structure learning into one optimization term, we can adequately capture the intrinsic geometric structure from missing and available data. In addition, the weight of incomplete views is learned adaptively to balance the importance of different views. Furthermore, we integrate representation learning and clustering processes into a unified framework so that the clustering result can be obtained directly and without the need for post-processing. Experiments performed on eight incomplete multi-view datasets demonstrate the effectiveness of the proposed LS-IMC compared to other current approaches.