Edge-Preserving Image Smoothing Constraint in Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) of Hyperspectral Data

Siewert Hugelier; Raffaele Vitale; Cyril Ruckebusch

doi:10.1177/0003702817735598

Edge-Preserving Image Smoothing Constraint in Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) of Hyperspectral Data

Appl Spectrosc. 2018 Mar;72(3):420-431. doi: 10.1177/0003702817735598. Epub 2017 Nov 14.

Authors

Siewert Hugelier¹, Raffaele Vitale^{1

2}, Cyril Ruckebusch¹

Affiliations

¹ 1 27022 Université de Lille, Sciences et Technologies, LASIR, CNRS, Lille, France.
² 2 KU Leuven, Molecular Imaging and Photonics Unit, Heverlee, Belgium.

PMID: 28922929
DOI: 10.1177/0003702817735598

Abstract

This article explores smoothing with edge-preserving properties as a spatial constraint for the resolution of hyperspectral images with multivariate curve resolution-alternating least squares (MCR-ALS). For each constrained component image (distribution map), irrelevant spatial details and noise are smoothed applying an L₁- or L₀-norm penalized least squares regression, highlighting in this way big changes in intensity of adjacent pixels. The feasibility of the constraint is demonstrated on three different case studies, in which the objects under investigation are spatially clearly defined, but have significant spectral overlap. This spectral overlap is detrimental for obtaining a good resolution and additional spatial information should be provided. The final results show that the spatial constraint enables better image (map) abstraction, artifact removal, and better interpretation of the results obtained, compared to a classical MCR-ALS analysis of hyperspectral images.

Keywords: Hyperspectral image analysis; MCR-ALS; constraint; edge preservation; multivariate curve resolution–alternating least squares; spatial smoothing.