Automated Phase Unwrapping in DHM with Deep Learning

Abstract

The digital holographic microscopy (DHM) can provide quantitative phase images related to the morphology and content of biological samples. This chapter introduces a new deep learning model that can effectively resolve the incomplete phase unwrapping in real-time. Convolutional neural network-based models face the problem of an abrupt phase shift that happens in the numerical phase unwrapping algorithms. It also introduces a generative adversarial network (GAN) to completely unwrap wrapped phase signals obtained using DHM which can automatically learn a proper adversarial loss function. The chapter aims to employ Pix2Pix GAN which consists of a generator and a discriminator and learns image-to-image translation with label images to automatically reconstruct unwrapped focused-phase images. Experimental results showed that the UnwrapGAN model could automatically reconstruct an unwrapped phase image in-focus from a wrapped phase image regardless of the reconstruction distance.