Multispectral Image Restoration via Inter- and Intra-block Sparse Estimation based on Physically-induced Joint...

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Traditional low-level vision algorithms prioritize spatial domain precision.
• High spectral domain precision is crucial for scientific applications like spectral analysis and classification.
• Existing methods often process spectral bands independently, limiting spatiospectral fidelity.

Purpose of the Study:

• To develop a new approach for multispectral image restoration that achieves high fidelity in both spatial and spectral domains.
• To introduce and exploit previously unexplored joint spatiospectral sparsities.
• To improve upon existing methods for superresolution and denoising of multispectral images.

Main Methods:

• Utilized a bidirectional image formation model to identify aligned spatial discontinuities across spectral bands.
• Developed a novel inter- and intra-block sparse estimation approach operating on 3D spatiospectral sample blocks.
• Combined intra-block sparsity (ℓ1,2-norm minimization) and inter-block low rank priors for robust regularization.

Main Results:

• Demonstrated the effectiveness of the new approach on superresolution and denoising tasks for multispectral images.
• Empirical results established the validity and advantages over current state-of-the-art methods.
• The joint spatiospectral sparsity prior significantly enhances image restoration quality.

Conclusions:

• The proposed method effectively restores multispectral images by exploiting joint spatiospectral properties.
• Processing 3D spatiospectral blocks offers superior performance compared to 2D patch-based methods.
• This work advances multispectral image restoration, offering significant improvements for scientific and technical applications.