Multiview Subspace Clustering by an Enhanced Tensor Nuclear Norm.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Tensor-singular value decomposition (t-SVD) based multiview subspace methods show promise but struggle with real-world noise and illumination variations.
• Current tensor-nuclear norm minimization (TNNM) in t-SVD treats singular values equally, which is suboptimal for tasks like coefficient matrix learning.

Purpose of the Study:

• To develop a more robust multiview subspace clustering method by addressing the limitations of existing t-SVD approaches.
• To introduce a weighted tensor Schatten p-norm minimization (WTSNM) framework that better exploits the differing importance of singular values.

Main Methods:

• Investigated the weighted tensor Schatten p-norm based on t-SVD.
• Developed an efficient algorithm to solve the weighted tensor Schatten p-norm minimization (WTSNM) problem.
• Integrated WTSNM for coefficient matrix learning into a unified framework with spectral clustering for multiview clustering.

Main Results:

• The proposed WTSNM method effectively handles noise and illumination changes in multiview data.
• The learned coefficient matrix captures both cluster structure and high-order information from multiview data.
• Extensive experiments demonstrated the superior efficiency of the proposed method across six evaluation metrics.

Conclusions:

• The novel multiview clustering method based on WTSNM offers significant improvements over traditional t-SVD approaches.
• This approach provides a more robust and effective solution for multiview subspace clustering in practical scenarios.
• The weighted norm effectively leverages the distinct information represented by different singular values.