Improved level set method for particle reconstruction from X-ray computed tomography images

Abstract

The heat and fluid flow properties of granular materials are largely controlled by their 3D microstructures. With recent advancements in image-based network modeling, heat transfer and fluid dynamic physics can be effectively simulated. Model accuracy depends on precise 3D reconstruction of particles from X-ray computed tomography (XRCT) images; however, existing workflows that combine watershed algorithms with level set methods require subjective manual inputs in preprocessing and lack controllability in optimization processes. To address these problems, this work proposes an automated workflow that can achieve high precision in the 3D reconstruction of particles. Oversegmentation instances by the traditional watershed algorithm are eliminated through morphological dilations, providing better inputs for the level set optimization process to lay a solid foundation for particle boundary extraction. The results of the proposed level set method are validated with paired multiresolution images of nearly spherical Ottawa and irregular angular sand. Compared with the current state-of-the-art method, the proposed method consistently and significantly obtains smoother particle boundaries and more accurate particle volumes and interparticle contact areas.