Limited-angle artifacts removal and jitter correction in soft x-ray tomography via physical model-driven deep learning

Abstract

Soft x-ray nanoscale tomography provides high-resolution three-dimensional visualization of the imaged objects and promotes the development of multiple research fields. However, the current challenges lie in the presence of limited-angle artifacts and projection jitter, which degrade the imaging resolution and quality. To address these issues, we propose a physical model-driven deep learning including forward and backward CT models. Combing with the iterative algorithm, the proposed method simultaneously suppresses the limited-angle and jitter artifacts. Furthermore, the physical model generates plenty of data to overcome the requirement of abundant experimental datasets. Both simulation and experiment demonstrate the feasibility and validity of the proposed reconstruction algorithm.