Abstract: Multi-view subspace clustering, as a clustering algorithm for multi-view data, aims to learn a consensus subspace for clustering. However, the existing multi-view clustering algorithms only focus on the original multi-view, ignoring the data obtained by direct feature concatenation. The algorithm proposed in this paper focuses on the mutual learning of the original multi-view and the feature concatenation view, and obtains a more suitable subspace representation through error reconstruction and structural constraint subspace. At the same time, the weight relationship between multi-view and feature concatenation view is also considered. Finally, experiments are conducted on four benchmark datasets to verify the effectiveness of the algorithm. 

Key words: multi-view subspace, consensus matrix, feature concatenation, structured constraint