Incomplete Multi-View Clustering Based on Weighted Adaptive Graph Learning

Abstract

Multi-view clustering methods utilize complementary and consistent information among different views to classify samples into correct clusters. However, traditional multi-view clustering methods are proposed based on complete datasets. In practical applications, complete data samples rarely exist, and incomplete data are more common. This paper proposes an incomplete multi-view clustering algorithm based on weighted adaptive graph learning. Specifically, we first introduce a distance regularization term and integrate it into the framework of low-rank representations to learn graphs with both local and global structure of the data. Then, we use spectral clustering to achieve a low dimensional matrix for each graph. Finally, we use a weighted fusion mechanism to learn a consensus representation, and utilize the K-means algorithm to get the final clustering results. Experimental results on different incomplete datasets demonstrate the effectiveness of the algorithms.