Hierarchical clustering of X-Ray Fluorescence images using a dissimilarity based on the likelihood ratio statistic
Abstract
The study of ancient materials is marked by their heterogeneity, uniqueness, and poor prior knowledge. Their analysis must be extensive, non-destructive, and precise. For these reasons, X-Ray Fluorescence (XRF) imaging is widely used for elemental analysis. X-Rays exposure time must balance extensiveness and sample preservation, a trade-off highly dependent on data exploitation: sufficient data must be collected for relevance without damaging the sample.
To address this, this article extends the work of Celeux et al. ( 2022) on spatially constrained hierarchical clustering of spectral images. It focuses on the probabilistic model characterizing XRF, deriving a dissimilarity and an associated cost function for comparing spectra. Tested within the aforementioned algorithm, it improves segmentation robustness to noise and relevance over the previously used χ 2 distance. Our results suggest equivalent material characterisation could be achieved with reduced exposure time, minimizing damage and enabling the collection of larger images.
Related
- Rate-Optimal Spectral Likelihood-Ratio Clustering in Stochastic Block Models
- Hierarchical Clustering Can Jointly Satisfy Richness, Consistency, and Scale Invariance
- Round-Trip KNN Clustering: multiscale hierarchical cluster detection on directed nearest-neighbour graphs
- Robust conditional dimension reduction for dissimilarity data
- Limits of Generalized Magnitudes of Dissimilarity Spaces
BibTeX
@misc{megel2026hierarchical,
title = {{Hierarchical clustering of X-Ray Fluorescence images using a dissimilarity based on the likelihood ratio statistic}},
author = {Maxime Megel and Gilles Celeux and Serge X. Cohen and Agnes Grimaud and Clara Hairie and Christine Keribin},
howpublished = {HAL},
year = {2026},
month = oct,
url = {https://hal.science/hal-05799747v2},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Hierarchical clustering of X-Ray Fluorescence images using a dissimilarity based on the likelihood ratio statistic
AU - Maxime Megel
AU - Gilles Celeux
AU - Serge X. Cohen
AU - Agnes Grimaud
AU - Clara Hairie
AU - Christine Keribin
PY - 2026
DA - 2026/10/14
PB - HAL
UR - https://hal.science/hal-05799747v2
AB - The study of ancient materials is marked by their heterogeneity, uniqueness, and poor prior knowledge. Their analysis must be extensive, non-destructive, and precise. For these reasons, X-Ray Fluorescence (XRF) imaging is widely used for elemental analysis. X-Rays exposure time must balance extensiveness and sample preservation, a trade-off highly dependent on data exploitation: sufficient data must be collected for relevance without damaging the sample. To address this, this article extends the work of Celeux et al. ( 2022) on spatially constrained hierarchical clustering of spectral images. It focuses on the probabilistic model characterizing XRF, deriving a dissimilarity and an associated cost function for comparing spectra. Tested within the aforementioned algorithm, it improves segmentation robustness to noise and relevance over the previously used χ 2 distance. Our results suggest equivalent material characterisation could be achieved with reduced exposure time, minimizing damage and enabling the collection of larger images.
ER -