Phase retrieval with a dual recursive scheme.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Optical sensors inherently lack phase information, necessitating phase retrieval (PR) for reconstructing wavefronts.
• Intensity measurements are common in optical imaging, but lack crucial phase data.
• Phase retrieval is vital for numerous applications, including microscopy and holography.

Purpose of the Study:

• To develop an advanced, learning-based method for robust phase retrieval.
• To enhance the accuracy and quality of reconstructed images from intensity data.
• To address the challenge of phase information loss in optical measurements.

Main Methods:

• Proposed a learning-based recursive dual alternating direction method of multipliers (RD-ADMM).
• Implemented a dual and recursive scheme, solving primal and dual problems separately.
• Developed a learning-based coded holographic coherent diffractive imaging system for automatic reference pattern generation.

Main Results:

• The RD-ADMM method demonstrated effectiveness and robustness in phase retrieval.
• Achieved higher-quality results compared to conventional PR methods, especially under high noise levels.
• Validated the approach on diverse image types, confirming its practical utility.

Conclusions:

• The proposed RD-ADMM is an efficient and robust solution for phase retrieval.
• The dual structure and learning-based approach significantly improve reconstruction quality.
• The method shows promise for advanced optical imaging applications requiring accurate phase information recovery.