Phase retrieval via nonlocal complex-domain sparsity.

Area of Science:

• Coherent Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Phase retrieval is crucial for coherent imaging systems but struggles with noise and limited exposure.
• Traditional algorithms often fail to reconstruct fine details accurately when noise is present.

Purpose of the Study:

• To develop a noise-robust phase retrieval framework for high-fidelity image reconstruction.
• To enhance the recovery of fine details in coherent imaging despite measurement noise.

Main Methods:

• An iterative framework incorporating nonlocal structural sparsity via low-rank regularization.
• Joint optimization of sparsity regularization and data fidelity using forward models.
• An adaptive iteration strategy to improve computational efficiency by adjusting matching frequency.

Main Results:

• The framework effectively suppresses artifacts caused by measurement noise.
• Satisfying detail recovery was achieved through joint optimization.
• Demonstrated effectiveness in coherent diffraction imaging and Fourier ptychography.
• Achieved approximately 7 dB higher peak signal-to-noise ratio (PSNR) compared to conventional methods.

Conclusions:

• The proposed iterative framework offers a noise-robust solution for phase retrieval.
• The technique significantly enhances image quality and detail recovery in coherent imaging.
• This approach provides a more efficient and accurate alternative to traditional phase retrieval algorithms.