Correntropy-Based Low-Rank Matrix Factorization With Constraint Graph Learning for Image Clustering.

Area of Science:

• Computer Vision
• Machine Learning
• Data Mining

Background:

• Image clustering is crucial for organizing large datasets.
• Existing methods struggle with outliers and effectively utilizing limited label information.
• Semisupervised learning offers a promising approach to enhance clustering performance.

Purpose of the Study:

• To propose a novel low-rank matrix factorization model for semisupervised image clustering.
• To enhance robustness against outliers using the maximum correntropy criterion (MCC).
• To leverage label information for adaptive local structure learning.

Main Methods:

• Developed a low-rank matrix factorization model incorporating the maximum correntropy criterion (MCC).
• Introduced a constraint graph learning framework to integrate label information.
• Employed an iterative algorithm based on Fenchel conjugate (FC) and block coordinate update (BCU) for model optimization.
• Analyzed the convergence properties of the proposed algorithm.

Main Results:

• The proposed model demonstrated improved performance on six real-world image datasets.
• Achieved superior clustering accuracy and mutual information compared to eight state-of-the-art methods.
• The algorithm exhibited both objective sequential convergence and iterate sequential convergence.

Conclusions:

• The novel semisupervised image clustering model effectively handles outliers and utilizes label information.
• The proposed method offers a robust and accurate solution for image clustering tasks.
• The developed iterative algorithm ensures reliable convergence and efficient computation.