Total variation regularization via continuation to recover compressed hyperspectral images

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral imaging generates large datasets, necessitating efficient compression and reconstruction techniques.
• Existing methods for compressed hyperspectral image data decoding often face challenges with complexity and reconstruction quality.
• The subgradient method offers a foundation for optimization but requires adaptation for hyperspectral data challenges.

Purpose of the Study:

• To develop a low-complexity decoding scheme for compressed hyperspectral image data.
• To enhance the efficiency and accuracy of hyperspectral image reconstruction algorithms.
• To introduce a method that exploits spectral smoothness for improved convergence rates.

Main Methods:

• Modified a total variation-based regularization problem with a residual constraint.
• Utilized convex optimality conditions to ensure problem equivalency.
• Developed a scheme incorporating spectral smoothness for informed starting points to accelerate convergence.
• Conducted numerical experiments on synthetic and real hyperspectral data.

Main Results:

• Demonstrated the effectiveness of the proposed reconstruction algorithm.
• Validated the method's capability in exploiting spectral smoothness.
• Showcased improvements in the rate of convergence through informed starting points.
• Provided evidence for enhanced reconstruction quality and reduced run times.

Conclusions:

• The proposed low-complexity scheme effectively decodes compressed hyperspectral image data.
• Exploiting spectral smoothness significantly improves algorithm performance and convergence.
• The method holds potential for advancing future compressed hyperspectral image reconstructions in terms of quality and efficiency.