A Non-Local Superpatch-Based Algorithm Exploiting Low Rank Prior for Restoration of Hyperspectral Images.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Hyperspectral images (HSIs) are crucial for various applications but are susceptible to degradation.
• Common degradations include noise (Gaussian, Poisson, mixed) and structural artifacts like dead lines and stripes.
• Existing restoration methods often require prior knowledge of the degradation, limiting their applicability.

Purpose of the Study:

• To develop a novel, robust algorithm for hyperspectral image restoration.
• To address the challenge of restoring HSIs without prior knowledge of degradation types.
• To improve the accuracy and reliability of HSI restoration.

Main Methods:

• Exploitation of spatial and spectral redundancies using superpatches.
• Formulation of a restoration algorithm with structural similarity index measure (SSIM) as the data fidelity term.
• Incorporation of nuclear norm as the regularization term for enhanced restoration.

Main Results:

• The algorithm successfully restores hyperspectral images corrupted by various noise types and artifacts.
• Demonstrated capability to recover spectral information even under severe degradation.
• Experimental results show superior performance compared to state-of-the-art low-rank methods.

Conclusions:

• The proposed algorithm offers a competitive and effective solution for hyperspectral image restoration.
• It provides a versatile approach applicable to diverse degradation scenarios.
• The method advances the field of HSI processing and analysis.