Anti-noise variational sparse Bayesian estimation ghost imaging based on 3Level factor graph.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Signal Processing

Background:

• Existing compressed sensing ghost imaging (CSGI) schemes face limitations in anti-noise performance and parameter setting.
• Accurate imaging of complex objects at low sampling rates remains a challenge.

Purpose of the Study:

• To propose an innovative Bayesian compressed sensing ghost imaging (BCSGI) method with superior anti-noise capabilities.
• To enhance reconstruction accuracy and imaging quality, particularly for complex objects under noisy conditions.
• To reduce computational time while maintaining high precision in CSGI.

Main Methods:

• Utilizing sparse representation via K-Singular Value Decomposition (KSVD).
• Implementing a 3Level (3L)-hierarchical variational message passing (VMP) algorithm.
• Applying Bayesian inference for compressed sensing reconstruction.

Main Results:

• The proposed method demonstrates superior anti-noise performance compared to existing CSGI techniques.
• Achieved accurate imaging of highly complex objects at low sampling rates (below 12.2%) with varying noise levels.
• Outperforms existing Bayesian compressive sensing ghost imaging (BCSGI) in reconstruction accuracy and imaging quality.
• Moderately reduces time consumption compared to BCSGI while ensuring high precision.

Conclusions:

• The innovative BCSGI method effectively overcomes limitations of parameter presetting in traditional CSGI.
• The approach offers significant improvements in imaging complex objects under noisy conditions and low sampling rates.
• This work presents potential applications for CSGI in biomedical imaging.