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.
Recently, we proposed a deep-learning (DL) -based method for solving coherent imaging inverse problems, known as coherent plug and play (CPnP). CPnP is a regularized inversion framework that works with coherent imaging data corrupted by phase errors. The algorithm jointly produces a focused and speckle-free image and an estimate of the phase errors. The algorithm combines physics-based propagation models with image models learned with DL and produces higher-quality estimates when compared to other non-DL methods. Previously, we were only able to demonstrate CPnP using simulated data. In this work, we design a coherent imaging test bed to validate CPnP using real data. We devise a method to obtain truth data for both the images and the phase errors. This allows us to quantify performance and compare different algorithms. Our results validate the improved performance of CPnP when compared to other existing methods.
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.
The alert did not successfully save. Please try again later.
Casey J. Pellizzari, Timothy J. Bate, Kevin P. Donnelly, Mark F. Spencer, "Solving coherent-imaging inverse problems using deep neural networks: an experimental demonstration," Proc. SPIE 12239, Unconventional Imaging and Adaptive Optics 2022, 122390A (4 October 2022); https://doi.org/10.1117/12.2633016