Augmenting Polarized Light Microscopy with Computational Imaging and Deep Learning for Cultural Heritage

FAIN: PR-284405-22

Northwestern University (Evanston, IL 60208-0001)
Marc Sebastian Walton (Project Director: May 2021 to March 2022)
Aggelos K. Katsaggelos (Project Director: March 2022 to present)
Aggelos K. Katsaggelos (Co Project Director: February 2022 to March 2022)
Oliver Strides Cossairt (Co Project Director: February 2022 to present)
Florian Willomitzer (Co Project Director: February 2022 to present)

The development of image-based, quantitative protocols for Polarized Light Microscopy (PLM) using hardware and deep learning algorithms to generate image data for pigment identification and diagnosis of patterns of deterioration.

The project proposed here builds on this significant infrastructure and know-how within the conservation profession on PLM use. Focusing on the extensive archive of pigment dispersion slides at the Art Institute of Chicago1 and the Forbes collection at Harvard Art Museums as source materials, this proposal aims to maximize the amount of information extracted from PLM through recent advances in sensor hardware combined with computational imaging and deep learning. In short, we will be modernizing PLM by "harnessing the data revolution"to provide cutting-edge resources for conservators to make pigment identifications and to diagnose patterns of deterioration. As a core part of our dissemination, we will be making both the data collected as well as software pipelines open source for use by anyone and accessible through the Center of Scientific Studies in the Arts'' (NU-ACCESS) online presence.

Associated Products

Polarized Light Microscope (Equipment)
Name: Polarized Light Microscope
Description: Setup for customized Polarized Light Microscope with LED array light source
Location: Northwestern University
Year: 2023

Simulation of Polarized LightMicroscopy for Multiple Analyzer Angles (Conference Paper/Presentation)
Title: Simulation of Polarized LightMicroscopy for Multiple Analyzer Angles
Author: Manuel Ballester
Author: Zoey Ho
Author: Asami Odate
Author: Marc Walton
Author: Aggelos Katsaggelos
Abstract: We have developed an efficient simulator for polarized light microscopy experiments. It supports calculations for multiple analyzer angles across different channels of a polarized camera, enhancing imaging capabilities.
Date: 9/9/2024
Primary URL: https://www.optica.org/events/topical_meetings/latin_america_optics_and_photonics_conference/
Primary URL Description: doi: 10.20944/preprints202409.0664.v1
Conference Name: Optica Latin America Optics and Photonics Conference

Towards quantitative polarized light microscopy with Fourier ptychography coupled with deep learning for cultural heritage (Conference Paper/Presentation)
Title: Towards quantitative polarized light microscopy with Fourier ptychography coupled with deep learning for cultural heritage
Author: Asami Odate
Author: Henry Chopp
Author: Manuel Ballester
Author: Zoey Ho
Author: Aggelos K Katsaggelos
Author: Marc Walton
Abstract: Polarized light microscopy (PLM) methods are widely used in conservation and are found in cultural institutions and material analysis labs worldwide. This technique, developed in the early 20th century and popularized by Walter McCrone's practical courses in the 1980s and 1990s, is an essential tool for characterization. Despite its widespread use, two major factors limit the broader adoption of PLM. Firstly, the analog operation of these microscopes does not make use of the wealth of micrograph data accumulated over the last 50 years. Secondly, PLM heavily relies on extensively trained microscopists, and the subjective nature of the process limits its use as a qualitative diagnostic.  To tackle these issues, our project aims to develop a database of PLM-derived images and create a fast image retrieval algorithm to access the closest matches to unknown specimens. Additionally, we aim to establish the PLM as a quantitative tool by leveraging Fourier ptychography (FP) combined with a physics-informed neural network (PINN). Both the quantification of PLM images and the database will ensure the robustness of PLM as a technique adaptable to the future of conservation. In this proceeding, we propose the hardware implementation of FP-PLM/PINN, the database organization structure, and identify issues associated with database development.
Date: 7/7/2024
Primary URL: https://www.grc.org/scientific-methods-in-cultural-heritage-research-conference/2024/
Primary URL Description: Conference webpage
Conference Name: Scientific Methods in Cultural Heritage Research, Gordon Research Conference

---