Correntropy-Induced Robust Low-Rank Hypergraph.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Hypergraph learning effectively models high-order data correlations in image processing.
• Existing methods are vulnerable to non-Gaussian noise, limiting performance.
• Robust hypergraph structure is crucial for accurate data correlation formulation.

Purpose of the Study:

• To develop a noise-resistant hypergraph learning model for improved image processing.
• To enhance robustness against various non-Gaussian noises.
• To improve performance in image clustering and semi-supervised image classification.

Main Methods:

• Constructing a hypergraph using low-rank representation to capture global data structure and group effects.
• Employing a correntropy-induced local metric to measure reconstruction errors, robust to non-Gaussian noise.
• Utilizing Frobenius-norm based regularization with a low-rank regularizer to manage singular values and select hyperedges/weights.

Main Results:

• The proposed model demonstrates superior robustness against non-Gaussian noises.
• Quantitative evaluations show significant performance enhancements on benchmark datasets.
• The method outperforms state-of-the-art hypergraph models in image clustering and semi-supervised classification.

Conclusions:

• The developed noise-resistant hypergraph learning model offers a robust solution for image processing tasks.
• The combination of low-rank representation and correntropy-induced metric effectively handles noise.
• This approach advances hypergraph learning applications in computer vision.