Block-based spectral image reconstruction for compressive spectral imaging using smoothness on graphs.

Area of Science:

• Computer Vision
• Signal Processing
• Data Science

Background:

• Compressive spectral imaging (CSI) aims to reconstruct high-dimensional spectral data from limited measurements.
• Existing CSI methods often rely on assumptions like sparsity or low-rank properties, which may not always hold.
• The accurate reconstruction of spectral images is crucial for various applications, including remote sensing and medical imaging.

Purpose of the Study:

• To develop a novel reconstruction method for compressive spectral imaging.
• To leverage graph-based smoothness priors for improved spectral image reconstruction.
• To infer graph structures directly from available data, enhancing method adaptability.

Main Methods:

• A novel reconstruction method is proposed based on the assumption of spectral image smoothness over a graph.
• Graph structures are inferred from a panchromatic image using a state-of-the-art graph learning technique.
• Solutions are obtained efficiently via closed-form expressions by solving multiple sparse linear systems in parallel.

Main Results:

• The proposed method demonstrates superior performance compared to traditional sparsity-based and total variation methods.
• It outperforms recent approaches utilizing low-rank minimization and deep-learning-based plug-and-play priors.
• Extensive simulations and experimental results validate the effectiveness of the graph-based reconstruction approach.

Conclusions:

• The developed graph-based reconstruction method offers an effective alternative for compressive spectral imaging.
• Inferring graph structures from data enhances the robustness and applicability of the method.
• This approach holds potential for integration with deep neural networks and covariance estimation techniques for future advancements.