Enhancing image resolution of soils by stochastic multiscale image fusion

• Published in: Geoderma 2018-03, Vol.314, p.138-145
• Main Authors: Karsanina, Marina V., Gerke, Kirill M., Skvortsova, Elena B., Ivanov, Andrey L., Mallants, Dirk
• Format: Article
• Language: English
• Subjects: 3D multiscale digital model; Correlation functions; Multiscale fusion; Soil structure; Stochastic reconstruction; X-ray computed tomography
• ISSN: 0016-7061; 1872-6259
• DOI: 10.1016/j.geoderma.2017.10.055

Soil structure defines major physical properties and biophysical functions of soils. Imaging soil structure using different 2/3D techniques is a routine methodology used by soil scientists. Still, for structured soils their spatial variability and hierarchical structure imposes a significant challenge for all imaging methods in terms of field-of-view and resolution trade-off. While creating a truly multiscale 3D digital model of soil is without question of utmost importance, there is currently no single imaging method that could potentially encompass all necessary relevant soil scales within a single image. In this paper, we tested for the first time an image fusion technique to produce a multiscale soil image based on separate images obtained with different spatial resolutions. The method is based on universal soil structure descriptors, i.e. spatial correlation functions, which were shown to be very useful in soil applications. Using a relatively simple 2D test case based on X-ray tomography (XCT) images at three different scales, we show the applicability of image fusion for soil images and solve a long standing problem of imaging resolution. In total we fused seven images into a single image: one 114μm resolution macroscale XCT image (porosity