Abstract

Deep learning is widely used for quantitative phase imaging (QPI), but is prone to cause spatial frequency bias in the reconstruction. In this paper, we propose a split-and-synthesis framework, which consists of two-stages training and takes the phase samples based on uniform illumination and structured illumination from transport of intensity equation (TIE) as inputs. We show that our framework is efficient to calibrate the spatial frequency bias for accurate phase retrieval.

© 2022 The Author(s)