Multi-view subspace clustering based on low-rank tensor and angular information(基于低秩张量和角度信息的多视图子空间聚类)
∶Existing tensor-based multi-view subspace clustering methods have achieved remarkable success. However, these methods generally suffer from issues such as equivalent regularization of singular values and suboptimal construction of similarity matrices, which limit their performance. To address these...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Zhejiang University Press
2025-05-01
|
| Series: | Zhejiang Daxue xuebao. Lixue ban |
| Online Access: | https://doi.org/10.3785/j.issn.1008-9497.2025.03.005 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | ∶Existing tensor-based multi-view subspace clustering methods have achieved remarkable success. However, these methods generally suffer from issues such as equivalent regularization of singular values and suboptimal construction of similarity matrices, which limit their performance. To address these challenges, a novel multi-view subspace clustering algorithm based on low-rank tensor and angular information (MSCLTAI) is proposed. First, in order to fully utilize the significant difference of singular values, a weighted tensor nuclear norm based on tensor singular value decomposition (t-SVD) is introduced to learn the coefficient matrices in multi-view subspace clustering. Meanwhile, low-rank constraints are imposed on the tensor superimposed by the coefficient matrices to explore higher-order relevant information in the multi-view data. Next, considering the consistency between different views and the importance of their angular information, the consistency regularization term and information fusion reduction strategy are designed to obtain more effective similarity matrices for spectral clustering. Finally, the effectiveness of the proposed algorithm is verified through comparisons with several state-of-the-art algorithms on five real-world datasets.∶尽管基于张量的多视图子空间聚类算法已取得显著成效，然而这些算法普遍存在奇异值等价正则化和构造的相似矩阵非最优等问题，影响聚类性能。为此，提出一种基于低秩张量和角度信息的多视图子空间聚类（MSCLTAI）算法。首先，为充分利用奇异值的显著差异，引入基于张量奇异值分解（t-SVD）的加权张量核范数学习多视图子空间聚类的系数矩阵。同时，对由系数矩阵叠加成的张量施加低秩约束，以探索多视图数据的相关高阶信息。其次，考虑不同视图之间的一致性和其角度信息的重要性，分别设计了一致性正则化项和角度信息融合约简策略，令获得的相似矩阵更有效，并用其进行谱聚类。最后，在5个真实数据集上与多种优秀算法进行了比较，验证了算法的有效性。 |
|---|---|
| ISSN: | 1008-9497 |