Adaptive Rank and Structured Sparsity Corrections for Hyperspectral Image Restoration

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Hyperspectral images (HSIs) are prone to mixed noise (Gaussian, impulse, deadlines, stripes), degrading processing accuracy.
• HSI restoration often relies on low-rank matrix recovery (LRMR), but current methods like nuclear norm approximation have limitations.
• Existing methods struggle to capture the structured sparsity of sparse noise effectively.

Purpose of the Study:

• To develop a novel method for hyperspectral image restoration that addresses limitations in current low-rank matrix recovery techniques.
• To introduce adaptive rank and structured sparsity corrections (ARSSC) for more accurate HSI noise reduction.
• To improve the preservation of spatial-spectral structure information in restored HSIs.

Main Methods:

• Proposed the Adaptive Rank and Structured Sparsity Corrections (ARSSC) method for HSI restoration.
• Introduced two convex regularizers: rank correction (RC) and structured sparsity correction (SSC).
• Employed an efficient alternative direction method of multipliers (ADMM) algorithm to solve the optimization problem.

Main Results:

• ARSSC achieved a tighter approximation of noise-free HSI low-rank structure.
• The method effectively promoted the structured sparsity of sparse noise.
• Experimental results on simulated and real datasets demonstrated superior mixed noise removal and spatial-spectral information preservation compared to state-of-the-art methods.

Conclusions:

• The proposed ARSSC method significantly enhances hyperspectral image restoration quality.
• ARSSC offers a more robust approach to handling mixed noise in HSIs.
• The method effectively preserves essential spatial-spectral characteristics, crucial for downstream applications.