Fourier phase retrieval with untrained generative prior

Liyuan Ma; Hongxia Wang; Ningyi Leng; Ziyang Yuan

doi:10.1117/12.2663043

12 April 2023 Fourier phase retrieval with untrained generative prior

Liyuan Ma, Hongxia Wang, Ningyi Leng, Ziyang Yuan

Proceedings Volume 12565, Conference on Infrared, Millimeter, Terahertz Waves and Applications (IMT2022); 1256537 (2023) https://doi.org/10.1117/12.2663043
Event: Conference on Infrared, Millimeter, Terahertz Waves and Applications (IMT2022), 2022, Shanghai, China

Abstract

Fourier phase retrieval (FPR) is to recover a signal from its Fourier magnitude measurement. Due to the ill-posedness of the problem, it is often necessary to introduce prior information. Recently, replacing hand-crafted priors with data-learned priors, such as deep generative priors, has received a lot of attention in inverse problems. Note that the reconstruction performance of trained generative priors relies on a large amount of training data, in this paper we solve FPR problem with an untrained generative network, which approximates the unknown signal only with a fixed seed. We propose an algorithm that combines the alternating direction method of multipliers (ADMM) with an untrained generative network. Specifically, we model the problem as a constrained optimization problem, the ADMM is adopted to solve it alternately, and then an untrained generative network is embedded into the iterative process to constrain the estimated signal. The effectiveness of the proposed algorithm is demonstrated through experiments on grayscale images. Both PSNR, SSIM, and the visual quality of the reconstructed images are superior to that of the state-of-the-art algorithms, especially when the measurement is incomplete.

Citation Download Citation

Liyuan Ma, Hongxia Wang, Ningyi Leng, and Ziyang Yuan "Fourier phase retrieval with untrained generative prior", Proc. SPIE 12565, Conference on Infrared, Millimeter, Terahertz Waves and Applications (IMT2022), 1256537 (12 April 2023); https://doi.org/10.1117/12.2663043

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available