Consistent affinity representation learning with dual low-rank constraints for multi-view subspace clustering

Multi-view clustering aims to achieve better accuracy of data clustering by leveraging complementary information embedded in multi-view data. How to learn a consistent clustering-friendly affinity repre-sentation matrix is a crucial issue. In this paper, we propose a consistent affinity representation learning method with dual low-rank constraints for multi-view subspace clustering. To be specific, for capturing the high-order correlations and global consensus among views, we collect the subspace representations of all views into a 3-order tensor, which is imposed with the tensor singular value decomposition (t-SVD) based tensor nuclear norm for achieving the low-rank recovery. Thus, we learn a consistent affinity matrix by fusing multiple subspace representations on the Grassmann manifold rather than handling them in the Euclidean space. In order to enhance the global cluster structure in the uniform subspace, the low-rank constraint is imposed on the consistent affinity matrix. Furthermore, the local geometric structure of the uniform subspace is encoded via graph regularization. The established model can be solved via the alternating direction method of multipliers algorithm (ADMM). Ultimately, the proposed method is experimentally validated to be superior to other state-of-the-art clustering algorithms.(c) 2022 Elsevier B.V. All rights reserved.