Fast reconstruction of hyperspectral images from coded acquisitions using a separability assumption.

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Hyperspectral imaging captures detailed spectral information across numerous bands.
• Reconstructing hyperspectral data-cubes can be computationally intensive and data-hungry.
• Existing methods often require extensive training data or scene-specific prior knowledge.

Purpose of the Study:

• To develop a fast and efficient algorithm for hyperspectral image reconstruction.
• To enable reconstruction from a minimal amount of data without requiring prior training.
• To leverage spatial-spectral correlations for improved reconstruction.

Main Methods:

• A novel reconstruction algorithm is presented for hyperspectral images.
• The method utilizes a dual disperser hyperspectral imager.
• It employs a separability assumption based on homogenous spectral regions and spatial-spectral correlations.

Main Results:

• The algorithm successfully reconstructs hyperspectral data-cubes from a small number of acquisitions.
• Reconstruction is achieved near instantaneously, demonstrating high speed.
• The method does not require any prior knowledge of the scene, showcasing its versatility.

Conclusions:

• The presented algorithm offers a fast, training-free, and data-efficient solution for hyperspectral image reconstruction.
• This approach simplifies the process and reduces computational demands.
• It holds potential for real-time applications in various fields utilizing hyperspectral imaging.