Skip to Main content Skip to Navigation

optimisation of automatic segmentation of granular fragmented materials

Abstract : The physical properties of granular materials on a macroscopic scale derive from their microstructures. The segmentation of CT-images of this type of material is the first step towards simulation and modeling but it is not a trivial task. However, the quality of those images is often affected by the presence of noise and reconstruction artefacts. Obtaining 3D structures that fit the reality requires an adapted filter, which can only be obtained by a complete analysis of the material.This adapted filter enhances each grain and the full structure of the material is obtained by segmentation. However, non-spherical, elongated or non-convex objects fail to be separated with classical methods. Moreover, grains are commonly fragmented due to external conditions. Grains are ground into multiple fragments of different shape and volume; those fragments drift from one another in the binder phase.In this thesis, a complete process chain is proposed to segment complex structures that can be acquired by CT-scan. The raw data is first filtered and processed, and statistical features are extracted such as the number of phases, the number of grains of each phase, the size distribution and spectral identification of the phases. A primary segmentation is performed to identify every connection between touching grains and is based on the watershed transform. A hierarchy is built on the obtained contours to eliminate over-segmentation. Reconstruction of grains from fragments is achieved using affinities that match the local thickness and the regularity of the interface.Typical CT-images are voluminous, and the study of granular materials requires efficient use of modern computing architectures. Studying the state-of-the-art and its application to 3D data has oriented our choice has allowed us to balance the quality of segmentation and the computing cost. A novel segmentation method allows for higher performances while improving the quality of the result. Finally, two new algorithms are proposed for the labeling of connected components and for the watershed transformation.
Document type :
Complete list of metadata

Cited literature [158 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Monday, October 22, 2018 - 3:22:18 PM
Last modification on : Wednesday, November 17, 2021 - 12:27:15 PM
Long-term archiving on: : Wednesday, January 23, 2019 - 2:54:58 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01900940, version 1


Théodore-Flavien Chabardes. optimisation of automatic segmentation of granular fragmented materials. Image Processing [eess.IV]. Université Paris sciences et lettres, 2018. English. ⟨NNT : 2018PSLEM012⟩. ⟨tel-01900940⟩



Record views


Files downloads