Fourier ptychographic reconstruction with denoising diffusion probabilistic models

Fourier ptychographic microscopy (FPM) is a promising computational imaging technique that can bypass the diffraction limit of the objective lens and achieve high-resolution, wide field-of-view imaging. The FPM setups firstly capture a series of low-resolution (LR) intensity images by angle-varied illumination and then reconstruction algorithms recover the high-resolution (HR) complex-valued object from the LR measurements. The image acquisition process commonly introduces noise, ultimately leading to degradation in the quality of the reconstruction results. In this paper, we report a noise-robust Fourier ptychographic reconstruction method that generates the HR complex-valued object estimation using the image priors specified by denoising diffusion probabilistic models (DDPM). First, the initial estimation of the HR complex-valued object is matched with an intermediate state in the Markov chain defined by DDPM. Then, the noisy initial solution is iteratively updated to a high-quality reconstruction result in the reverse process of DDPM and gradient descent correction is incorporated to enforce data consistency with the LR measurements. The proposed method integrates DDPM specified image priors and gradient descent correction, achieving solutions with less noise-related artifacts and high fidelity for HR complex-valued object estimation in Fourier ptychographic reconstruction. We apply the proposed method on both synthetic and real captured data. The experimental results show that our method can efficiently suppress the impact of noise and improve reconstruction results quality.