Conceptualization of seeded region growing by pixels aggregation. Part 4: Simple, generic and robust extraction of grains in granular materials obtained by X-ray tomography

TL;DR

This paper introduces a robust method for extracting individual grains from 3D X-ray tomography images of granular materials, combining segmentation and splitting techniques to improve accuracy in complex samples.

Contribution

A simple, generic, and robust grain extraction method that effectively separates connected grains in X-ray tomography images, improving analysis of granular materials.

Findings

Effective separation of connected grains in complex samples

Good agreement with experimental data for property prediction

Applicable to various porous media and granular materials

Abstract

This paper proposes a simple, generic and robust method to extract the grains from experimental tridimensionnal images of granular materials obtained by X-ray tomography. This extraction has two steps: segmentation and splitting. For the segmentation step, if there is a sufficient contrast between the different components, a classical threshold procedure followed by a succession of morphological filters can be applied. If not, and if the boundary needs to be localized precisely, a watershed transformation controlled by labels is applied. The basement of this transformation is to localize a label included in the component and another label in the component complementary. A "soft" threshold following by an opening is applied on the initial image to localize a label in a component. For any segmentation procedure, the visualisation shows a problem: some groups of two grains, close one to each other, become connected. So if a classical cluster procedure is applied on the segmented binary image, these numerical connected grains are considered as a single grain. To overcome this problem, we applied a procedure introduced by L. Vincent in 1993. This grains extraction is tested for various complexes porous media and granular material, to predict various properties (diffusion, electrical conductivity, deformation field) in a good agreement with experiment data.