Dual structure-aware consensus graph learning for incomplete multi-view clustering

Compared to single-view data, multi-view data encompasses both additional complementary information and redundancies. The discriminative information presented in these aligned multiple views is helpful for enhancing the performance of clustering tasks. In reality, data views are frequently incomplete, which poses a significant challenge to the clustering task. In this paper, we introduce a new method, which we called Structured-aware Consensus Graph Learning for Incomplete Multi-View Clustering (SWCGLIMVC) to tackle the problem of incomplete multi-view clustering (IMVC). Specifically, considering that the neighbor relationships between samples are of utmost importance in unsupervised clustering tasks, SWCGLIMVC leverages the intrinsic geometry structure information of all samples and preserves their neighbor relationships through the graph Laplacian regularization constraint. Moreover, to reduce the adverse effects of the imbalanced useful information contained in different views, SWCGLIMVC incorporates a dynamically learnable vector to constrain the learning models of different views. This allows the method to effectively explore the information from all incomplete views for data clustering tasks. The effectiveness of SWCGLIMVC is evaluated by conducting experiments on six widely known datasets with the comparison of several state-of-the-art clustering methods. The experimental results show that the superior performance of SWCGLIMVC on IMVC tasks.