Artificial Intelligence in Digital Holographic Imaging

9Automated Phase Unwrapping in DHM with Deep Learning

9.1 Introduction

The digital holographic microscopy (DHM) can provide quantitative phase images related to the morphology and content of biological samples. Phase values in the reconstructed image are limited between −π and π. Thus, discontinuity may occur due to modulo 2π operation. A phase unwrapping process must be carried out to remove 2π phase discontinuities in the phase image and estimate the true continuous phase image. Phase unwrapping consists of finding the location of the phase jump and connecting adjacent pixels by adding or subtracting multiples of 2π to remove phase discontinuities.

Many phase unwrapping algorithms have been studied to solve challenging problems such as phase discontinuities. Advanced phase unwrapping algorithms can be divided into three types: global, region, and path‐following algorithms. Global algorithms can minimize differences between discrete gradients of wrapped and unwrapped phase images. Although these algorithms are robust, their computational requirements are large. Hence, they are unsuitable for real‐time live‐cell imaging applications. Region algorithms can split an image into smaller ones, unwrap regions with respect to each other, and merge them into larger regions. These algorithms have been regarded as a compromise between robustness and computational intensiveness. Region algorithms are further categorized into region‐based algorithms and tile‐based algorithms according ...