Graph Regularized Non-Negative Low-Rank Matrix Factorization for Image Clustering.

Area of Science:

• Machine Learning
• Computer Vision
• Data Mining

Background:

• Non-negative matrix factorization (NMF) is a common feature learning technique.
• Existing NMF methods often overlook that essential image information resides in low-rank components.
• Direct NMF application on high-dimensional data may not yield optimal representations.

Purpose of the Study:

• To propose a non-negative low-rank matrix factorization (NLMF) method for effective image clustering.
• To enhance robustness by incorporating manifold structure information via graph regularization.
• To develop an efficient algorithm for learning low-dimensional representations for clustering.

Main Methods:

• Developed a novel non-negative low-rank matrix factorization (NLMF) approach.
• Introduced graph regularization to NLMF to leverage manifold structure.
• Implemented an alternating iterative algorithm for optimization.
• Explored integration with robust principal component analysis.

Main Results:

• The proposed NLMF methods demonstrate superior performance in image clustering tasks.
• Graph regularization significantly improves robustness for data with manifold structures.
• Experimental results on four datasets confirm the effectiveness compared to existing methods.

Conclusions:

• The proposed graph-regularized NLMF is an effective technique for image clustering.
• Leveraging low-rank properties and manifold structures enhances feature representation.
• The developed methods offer a robust and efficient solution for image analysis.