Solution to the shearing problem via guided diffusion.

Area of Science:

• Optical Metrology
• Computational Imaging
• Wavefront Sensing

Background:

• Wavefront reconstruction from lateral shearing interferometry (LSI) is an ill-posed inverse problem.
• Existing methods often require multiple measurements or rely on potentially biased assumptions.
• Noise sensitivity and information loss are significant challenges in LSI.

Purpose of the Study:

• To introduce a novel framework, pseudoinverse-guided diffusion models (ΠGDM), for wavefront reconstruction.
• To leverage a pre-trained diffusion model as a data-driven prior for improved reconstruction.
• To address the limitations of conventional LSI techniques by enabling high-fidelity reconstruction from single shearograms.

Main Methods:

• Integration of LSI physics using Vector-Jacobian product guidance within a diffusion model framework.
• Utilizing a pre-trained unconditional diffusion model as a robust prior.
• Benchmarking against classical Zonal integration, Fourier, Tikhonov regularization, and supervised deep learning methods.

Main Results:

• ΠGDM demonstrates superior fidelity and noise robustness compared to benchmark methods in both simulations and experimental validation.
• The framework successfully recovers missing spatial frequencies while ensuring measurement consistency.
• Achieves high-precision metrology from single shearograms, a significant improvement over traditional approaches.

Conclusions:

• ΠGDM presents a novel and effective approach for wavefront reconstruction in LSI.
• It offers a versatile, zero-shot alternative to supervised learning methods, eliminating the need for task-specific training.
• The method enhances metrology precision and robustness, overcoming key challenges in LSI.