基于低秩张量图学习的不完整多视角聚类

Conventional multi-view clustering methods all assume that the give multi-view data is complete, i.e., all views are fully observed, which are not applicable to the incomplete multi-view clustering case with missing-views. To address this issue, we propose a method, called low-rank tensor graph learning (LASAR). To recover the missing connections corresponding to the missing views in the graph space, the proposed method integrates the low-rank tensor graph constraint and intra-view graph constraint into the multi-view spectral clustering framework. By exploring the inter-view information and intra-view information with respect to the observed views simultaneously, the proposed method can obtain the optimal completed graphs of all views and the optimal clustering indicator matrix shared by all views. Experimental results on five datasets with different missing-view rates show that the proposed method obtains better performance than 12 state-of-the-art incomplete multi-view clustering methods.