Graph-regularized tensor robust principal component analysis for hyperspectral image denoising.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral images (HSIs) are susceptible to noise, which can degrade image quality and hinder subsequent analysis.
• Traditional denoising methods may struggle to preserve the rich spectral and spatial information present in HSIs.
• The high-dimensional nature of HSIs presents unique challenges for noise reduction techniques.

Purpose of the Study:

• To develop a novel and effective denoising model for hyperspectral images.
• To enhance the accuracy of HSI denoising by preserving local geometric structures.
• To introduce an efficient algorithm for solving the proposed denoising model.

Main Methods:

• Developed a novel model named graph-regularized tensor robust principal component analysis (GTRPCA).
• Incorporated spectral graph regularization into tensor robust principal component analysis (TRPCA) to preserve local geometric structures.
• Utilized tensor singular value decomposition (t-SVD) and a tensor-based alternating direction method of multipliers (ADMM) algorithm for model solution.

Main Results:

• The proposed GTRPCA model demonstrated superior performance in denoising hyperspectral images.
• Preservation of local geometric structures led to more accurate noise removal.
• Experiments on synthetic and real datasets validated the effectiveness of the GTRPCA method.

Conclusions:

• GTRPCA is an effective approach for hyperspectral image denoising.
• The integration of spectral graph regularization significantly improves denoising accuracy.
• The developed ADMM algorithm efficiently solves the GTRPCA model for practical applications.