Newsletter 01/2025

By Hilda Deborah

Hyperspectral imaging (HSI) is an imaging technology that captures and analyzes across a wide range of wavelengths beyond what we humans can see. When you have the usual color camera, you have three values for each pixel, each representing the red, green, and blue channels. If you see the illustration above, HSI gives you a signal of typically hundreds of values per pixel. These signals can be used to, e.g., identify materials because they represent how a material interacts with light. 

In my research, among other things, I apply HSI to analyze paintings and their pigments, i.e., the actual coloring substance of paint. Why is it important to know the material of, e.g., a yellow paint? Well, because, depending on their composition, a yellow pigment will have a different degradation behavior over time. One might fade, while another might darken. Imagine if you have to retouch a painting but use an incorrect pigment for it; over the years, this will alter the appearance of the painting.

Some of my research works in applying HSI for paintings and a map collection are:

• H. Deborah, S. George, J. Y. Hardeberg. "Pigment mapping of The Scream (1893) based on hyperspectral imaging." ICISP 2014.

• J. D. Cutajar, A. Babini, H. Deborah, et al. "Hyperspectral imaging analyses of cleaning tests on Edvard Munch's monumental Aula paintings." Studies in Conservation, 2022.

• H. Deborah, C. Palandri, G. Oretti. "Estimating the Color Palette of Ortelius' Atlas: A case study of hyperspectral imaging for rapid pigment screening." IEEE Whispers 2023.

The Scandinavian Conference on Image Analysis (SCIA) 2025 will be held from June 23-25 at the University of Iceland, Reykjavík, Iceland. SCIA is a biennial scientific conference organized by the national pattern recognition societies in the Nordic countries (NOBIM for Norway) and is officially sponsored by the International Association of Pattern Recognition (IAPR).

Many NOBIM members participated in the conference; below are some of their contributions (members' names are written in bold). 

• S. A. Salahuddin, E. Wetzer, K. Wickstrøm, et al. "Assessing the efficacy of multi-task learning in mammographic density classification: A study on class imbalance and model performance."

• C. Salomonsen, K. Wickstrøm, E. Wetzer, S. Kuttner. "Physics-informed deep learning for improved input function estimation in motion-blurred dynamic [18F]FDG PET images."

• M. H. Zafar and J. Y. Hardeberg: Aligning subjective and objective assessments in super-resolution models.