Hyperspectral Image Restoration Using Weighted Group Sparsity-Regularized Low-Rank Tensor Decomposition.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Hyperspectral imagery (HSI) is susceptible to various noise types (Gaussian, impulse, stripe, deadline).
• Noise significantly degrades HSI visual quality and hinders accurate downstream processing.
• Traditional total variation (TV) regularization methods struggle to maintain shared group sparsity in spatial difference images across spectral bands.

Purpose of the Study:

• To develop an advanced HSI restoration technique that overcomes limitations of existing TV-regularized methods.
• To effectively remove mixed noise while preserving crucial spatial-spectral information in HSI data.
• To introduce a novel regularization strategy that leverages group sparsity for improved HSI denoising.

Main Methods:

• Proposed a weighted group sparsity regularization using the l2,1-norm on spatial difference images.
• Integrated low-rank Tucker decomposition to capture global spatial-spectral correlations.
• Developed a weighted group sparsity-regularized low-rank tensor decomposition (LRTDGS) model.
• Employed an augmented Lagrange multiplier algorithm for efficient model solving.

Main Results:

• The LRTDGS method demonstrated superior performance in HSI restoration compared to state-of-the-art techniques.
• Experimental results on simulated and real HSI data validated the effectiveness of the proposed approach.
• The method successfully removed mixed noise while preserving image details and spatial-spectral characteristics.

Conclusions:

• The proposed weighted group sparsity-regularized low-rank tensor decomposition (LRTDGS) offers an effective solution for HSI restoration.
• This method significantly enhances HSI quality by addressing noise contamination and preserving structural information.
• LRTDGS represents a valuable advancement in hyperspectral image denoising and restoration techniques.