Hyperspectral Image Denoising Using Nonconvex Local Low-Rank and Sparse Separation With Spatial-Spectral Total...

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral imaging (HSI) generates complex data with potential noise.
• Robust Principal Component Analysis (RPCA) is a technique used for HSI denoising.
• Existing RPCA methods may have limitations in accurately approximating low-rank and sparse components.

Purpose of the Study:

• To propose a novel nonconvex RPCA approach for improved HSI denoising.
• To develop more accurate approximations for rank and column-wise sparsity in HSI components.
• To enhance the spatial and spectral consistency of denoised HSIs.

Main Methods:

• Utilizing a log-determinant rank approximation for the low-rank component.
• Introducing a novel l2,log norm for column-wise sparsity approximation.
• Developing an efficient, closed-form solution: the l2,log-shrinkage operator.
• Incorporating spatial-spectral total variation regularization into the nonconvex RPCA model.

Main Results:

• The proposed method effectively denoises hyperspectral images.
• The novel l2,log-shrinkage operator provides an efficient solution for column-wise sparsity.
• The log-based nonconvex RPCA model with spatial-spectral total variation enhances HSI quality.
• Experiments on simulated and real HSIs validate the method's effectiveness.

Conclusions:

• The proposed nonconvex RPCA approach offers a significant advancement in HSI denoising.
• The developed l2,log norm and shrinkage operator are valuable tools for sparsity-based problems.
• The integration of spatial-spectral total variation improves the global smoothness and spectral consistency of recovered HSIs.