Haar Nuclear Norms With Applications to Remote Sensing Imagery Restoration.

Area of Science:

• Geospatial Science
• Image Processing
• Computer Vision

Background:

• Remote sensing image restoration commonly uses low-rank models.
• Existing methods struggle to balance low-rankness, local smoothness, and computational efficiency.
• Jointly modeling low-rank and local smoothness priors is challenging due to component entanglement.

Purpose of the Study:

• Propose a novel regularization method, the Haar Nuclear Norm (HNN), for efficient and effective remote sensing image restoration.
• Address limitations of existing methods in capturing both coarse structures and fine textures.
• Improve computational complexity and performance in image restoration tasks.

Main Methods:

• Developed the Haar Nuclear Norm (HNN) regularization technique.
• Transformed images into wavelet coefficients to separate frequency components.
• Applied nuclear norms to mode-3 unfolding matrices of wavelet coefficients for low-rank enforcement.

Main Results:

• HNN demonstrated effectiveness in hyperspectral image inpainting, multi-temporal cloud removal, and denoising.
• Achieved 1-4 dB performance improvement in inpainting tasks compared to state-of-the-art methods.
• Provided a 10-28x speedup for inpainting tasks, showcasing significant computational efficiency.

Conclusions:

• The Haar Nuclear Norm (HNN) offers a superior approach to remote sensing image restoration.
• HNN effectively balances the recovery of structural and textural information.
• The proposed method presents a computationally efficient and high-performing solution for various restoration challenges.