A Denoising Method for X-Ray Ptychography Combining a Physical Diffraction Model With a Deep Image Prior Network

X-ray ptychography is a lensless imaging technology with promising applications that can achieve nanometer resolution. However, the noise in the diffraction patterns degrades the performance of the phase recovery algorithms for ptychography. In reconstructed objects, the artifacts can affect details. We categorize the noise in the diffraction patterns into static intensity (SI) and dynamic random noise (DRN), which lead to periodic artifacts (PAs) and random noise in the reconstructed object. A denoising method for X-ray ptychography is, therefore, proposed to suppress the SI and DRN by a unified mathematical model. An explicit noise constraint based on the physical diffraction model (PDM) and an implicit prior regarding the object for network extraction are integrated to construct a joint noise constraint (JNC). Simulations and soft X-ray experiments demonstrate the advanced capability of the proposed method for noise suppression. Compared with other methods [extended ptychographical iterative engine (ePIE) and periodic-artifact suppressing algorithm (PASA)], the proposed method has high robustness and generality.