Multiple Complementary Priors for Multispectral Image Compressive Sensing Reconstruction.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Compressive sensing (CS) is crucial for multispectral imagery (MSI) reconstruction.
• Nonlocal-based tensor methods leverage MSI's nonlocal self-similarity (NSS) but neglect external image priors.
• Existing methods often produce ringing artifacts due to patch aggregation.

Purpose of the Study:

• To propose a novel Multiple Complementary Priors (MCP) approach for highly effective MSI-CS reconstruction.
• To integrate diverse image priors, including internal/external and shallow/deep, for improved reconstruction quality.
• To address limitations of existing methods by incorporating deep image priors and reducing artifacts.

Main Methods:

• The proposed MCP method jointly exploits nonlocal low-rank and deep image priors.
• A hybrid plug-and-play framework incorporates multiple complementary priors.
• An alternating direction method of multipliers (ADMM) algorithm is developed for optimization.

Main Results:

• The MCP algorithm demonstrates superior performance compared to state-of-the-art CS techniques for MSI reconstruction.
• Experimental results validate the effectiveness of integrating multiple complementary priors.
• The method successfully mitigates ringing artifacts often seen in other approaches.

Conclusions:

• The novel MCP approach offers a significant advancement in MSI-CS reconstruction.
• Combining diverse image priors, particularly deep priors, enhances reconstruction accuracy and quality.
• The developed ADMM-based optimization framework provides a tractable solution for the complex reconstruction problem.