Plug-and-Play Priors for Multi-Shot Compressive Hyperspectral Imaging.

Area of Science:

• Optics and Photonics
• Image Processing
• Computational Imaging

Background:

• Multi-shot coded aperture snapshot spectral imaging (CASSI) captures hyperspectral images (HSI) using multiple snapshots.
• Existing CASSI reconstruction algorithms often lack effectiveness and flexibility, especially for spectrally rich scenes.
• The number of measurements in CASSI increases with snapshots, demanding robust reconstruction techniques.

Purpose of the Study:

• To develop a flexible and effective reconstruction method for multi-shot CASSI systems.
• To leverage advanced denoising priors to improve hyperspectral image recovery.
• To address the limitations of current reconstruction algorithms in CASSI.

Main Methods:

• Proposed a plug-and-play (PnP) method utilizing denoisers as priors for multi-shot CASSI reconstruction.
• Implemented the PnP method based on the primal-dual algorithm with linesearch (PDAL) for flexibility across CASSI mechanisms.
• Introduced a novel subspace-based nonlocal reweighted low-rank (SNRL) denoiser to exploit HSI spectral and spatial correlations.

Main Results:

• Successfully recovered hyperspectral images from two snapshots with mean peak signal-to-noise ratio (MPSNR) above 50 dB.
• Demonstrated high-quality reconstruction of a 512x512x31 balloon dataset from the CAVE dataset.
• Achieved significant performance improvements compared to existing state-of-the-art CASSI reconstruction methods.

Conclusions:

• The proposed PnP-PDAL method with the SNRL denoiser offers a flexible and effective solution for multi-shot CASSI.
• Integrating advanced denoising priors significantly enhances hyperspectral image reconstruction quality.
• The method shows strong potential for applications requiring detailed spatial and spectral information from compressed measurements.