Correspondence imaging using mixed illumination patterns in complex environments with random disturbances.

Area of Science:

• Optics
• Computational Imaging
• Machine Learning

Background:

• Single-pixel detection optical imaging struggles with complex environments due to single illumination patterns.
• Robustness and high-quality reconstruction are critical for advanced optical imaging applications.

Purpose of the Study:

• To develop a correspondence imaging (CI) method using mixed illumination patterns for improved object reconstruction.
• To enhance the robustness of optical imaging in complex environments with random disturbances.

Main Methods:

• Utilized a combination of sinusoidal and random illumination patterns in a correspondence imaging setup.
• Estimated and corrected dynamic scaling factors caused by environmental disturbances using random pattern data.
• Employed a physics-enhanced neural network (PENet) for final object image recovery without prior training data.

Main Results:

• Achieved high-fidelity object image reconstruction in optical experiments.
• Demonstrated significant robustness against random disturbances in complex environments.
• Validated the effectiveness of mixed illumination patterns and disturbance correction.

Conclusions:

• The proposed correspondence imaging method with mixed illumination patterns offers high performance in challenging optical imaging scenarios.
• This approach provides a pathway for developing robust and high-fidelity CI systems for complex environments.
• The physics-enhanced neural network approach enables data-free image recovery.