Two-frame randomly phase-shifted noisy interferograms phase retrieval based on intermediate frame generation network

Area of Science:

• Optical Metrology
• Image Processing
• Computational Physics

Background:

• Accurate phase retrieval is essential for interferometry, but traditional methods struggle with noise and physical process limitations.
• Existing two-frame phase retrieval techniques lack the precision needed for complex interferometric applications.

Purpose of the Study:

• To develop a novel, accurate, and robust method for phase retrieval from two-frame randomly phase-shifted noisy interferograms.
• To improve upon the limitations of traditional phase retrieval methods in interferometry.

Main Methods:

• Proposed an intermediate frame generation network (IFGNet) to process two-frame interferograms.
• IFGNet generates denoised inputs and intermediate frames, incorporating a multi-channel attention mechanism for physical information leverage.
• Phase retrieval performed using the Stoilov method on the five output frames, enhanced by a target-based training strategy (TBTS).

Main Results:

• The IFGNet method demonstrated superior accuracy and robustness compared to existing two-frame phase retrieval techniques.
• The multi-channel attention mechanism effectively predicted noise-free interferograms.
• Simulations and experiments validated the enhanced performance of the proposed method.

Conclusions:

• The novel IFGNet method offers a significant advancement in phase retrieval from noisy interferograms.
• The technique shows strong potential for practical applications in interferometry requiring high accuracy.
• The study highlights the effectiveness of deep learning and attention mechanisms in optical metrology.