Unified Framework for Faster Clustering via Joint Schatten p-Norm Factorization With Optimal Mean

Area of Science:

• Computer Vision
• Machine Learning
• Data Mining

Background:

• Subspace clustering is crucial for visual tasks but faces challenges in effectiveness and efficiency.
• Existing low-rank representation (LRR) methods often involve complex computations and biased estimations.
• The optimal mean embedded in LRR frameworks can hinder performance and increase computational load.

Purpose of the Study:

• To develop a novel, efficient, and effective subspace clustering approach for visual tasks.
• To address limitations of current LRR methods, including computational complexity and biased estimation.
• To introduce a unified framework that enhances performance and reduces processing time.

Main Methods:

• Proposed a nonconvex subspace clustering method via joint Schatten p-norm factorization with optimal mean (JS p NFOM).
• Employed tractable and scalable factor techniques to manage large-scale coefficient matrices.
• Utilized iterative optimization with multivariate weighting algorithms, avoiding singular value decomposition (SVD).

Main Results:

• Achieved faster nonconvex subspace clustering with improved effectiveness and efficiency.
• Demonstrated reduced computational complexity, especially for large datasets.
• Experimental results confirmed superior performance compared to state-of-the-art methods on public databases.

Conclusions:

• The proposed JS p NFOM framework offers a significant advancement in subspace clustering for visual data analysis.
• The method provides a scalable and computationally efficient solution without compromising accuracy.
• Theoretical convergence analysis and practical experiments validate its applicability in real-world scenarios.