Accurate phase aberration compensation with convolutional neural network PACUnet3+ in digital holographic microscopy

Abstract

To remove the effects of aberrations on quantitative phase image in digital holographic microscopy, a precise method that based on a data-driven phase aberrations compensations Unet3+ network (PACUnet3+) is presented. Unlike the existing methods that background segmentation should be implemented either in a phase image or in a sample hologram, the presented method doesn't need any preprocessing techniques. The PACUnet3+, which learning mappings between sample holograms and sample free holograms, produces a sample free hologram directly that contains all the background phase aberrations. Similar with double-exposure method, the phase aberrations can be removed by subtracting the background phase encoded in the generated sample-free hologram without any assumptions on the phase aberrations. The high fidelity of the method is demonstrated by measurement of engineered binary phase patterns. In phase aberrations correction, both numerical and experimental results confirmed the accuracy of the proposed method is benchmarking against the traditional double-exposure method.