Deep Learning based digital hologram processing

Li, Huayu (2021) Deep Learning based digital hologram processing. Masters thesis, Northern Arizona University.

Text Li_2021_deep_learning_based_digital_hologram_processing.pdf - Published Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Digital holography imaging utilizes the special properties of light and photons to capture the detailed images of microscopic objects, such as organs, tissues, and cells. Usually, phase information is used to reconstruct the surface of the target objects. Digital in-line holography is a widely used phase imaging method due to its simplicity of experimental setups. However, reconstructing the phase and amplitude planes from the raw holograms is hindered by couples of problems, such as the twin image problem due to the diffusion pattern and the noise caused by the environmental factors. Therefore, different analytical and numerical methods were proposed to tackle these problems. Recently, Deep Learning (DL) has been deployed in a range of computer vision tasks that achieve better performance than traditional machine learning methods. This thesis demonstrates the usage of DL methods to address hologram reconstruction and phase and amplitude image denoising. This thesis is formed in two aspects: (1) In Chapter 1, a deep learning based hologram reconstruction method was proposed that can reconstruct the phase and amplitude image from a single shot hologram without massive training data. We train a convolutional autoencoder with a physics-driven loss function that can produce twin image-free results from the captured hologram. We compared the proposed method with the state-of-art compressive sensing method on simulated data. We also compared the proposed method with the multi-height Transport of Intensity Equation (TIE) based algorithm on real-world optical experiments. Experimental results show that the proposed method could produce better results than single shot compressive sensing method and achieve comparable results to the multi-height TIE based algorithm. (2) Chapter 2 aims to boost the performance of Channel attention model for image denoising tasks. We proposed to use Discrete Wavelet Transform to decompose the feature maps into different frequency subbands and deploy the channel attention to these subbands. We evaluate the proposed methods on both natural image denoising and phase and amplitude image denoising. The results show that the proposed method achieves an outstanding performance for various noise and data types.

Item Type:	Thesis (Masters)
Publisher’s Statement:	© Copyright is held by the author. Digital access to this material is made possible by the Cline Library, Northern Arizona University. Further transmission, reproduction or presentation of protected items is prohibited except with permission of the author.
Keywords:	Holography; Digital imaging; Deep learning; Denoising;
Subjects:	T Technology > TA Engineering (General). Civil engineering (General)
NAU Depositing Author Academic Status:	Student
Department/Unit:	Graduate College > Theses and Dissertations College of Engineering, Informatics, and Applied Sciences > School of Informatics, Computing, and Cyber Systems
Date Deposited:	09 Feb 2022 17:57
Last Modified:	09 Feb 2022 17:57
URI:	https://openknowledge.nau.edu/id/eprint/5682

Actions (login required)

IR Staff Record View

Downloads