Dual-camera compressive hyperspectral imaging based on deep image prior and a guided filter.

Area of Science:

• Optics and Photonics
• Image Processing
• Computational Imaging

Background:

• Coded aperture snapshot spectral imaging (CASSI) captures 3D spectral data cubes on 2D sensors in one snapshot.
• CASSI reconstruction is often ill-posed, leading to suboptimal results.
• Integrating additional information, like panchromatic measurements, can improve reconstruction quality.

Purpose of the Study:

• To develop a dual-camera hyperspectral reconstruction method for CASSI.
• To leverage deep image prior (DIP) and guided filtering for enhanced spectral data reconstruction.
• To achieve high spatial and spectral resolution in reconstructed hyperspectral images without training data.

Main Methods:

• A dual-camera system combining CASSI and panchromatic measurements.
• Utilizing deep image prior (DIP) for self-supervised network learning.
• Employing iterative reconstruction with DIP and optimizing with guided filtering.

Main Results:

• The proposed method successfully reconstructs hyperspectral data.
• Achieved high spectral accuracy and spatial resolution in experimental results.
• Demonstrated effective reconstruction without the need for pre-existing training datasets.

Conclusions:

• The dual-camera approach with DIP and guided filtering significantly enhances CASSI reconstruction.
• This method provides a robust solution for acquiring high-fidelity hyperspectral data.
• The technique offers a training-data-free approach for spectral imaging applications.