Segmentation of X-Ray Computed Tomography Images: A Crucial Step for Quantitative Analysis of Porous Structures.

Nondestructive imaging methods such as X-Ray Computed Tomography (CT) yield high-resolution, grayscale representations of media pore space that can be discretisized and used in conjunction with advanced simulation techniques to model fluid distribution and flow and transport processes in porous media. Image discretization/binarization is the most crucial step affecting all subsequent quantitative analysis and modeling efforts. In order to obtain a morphological representation of the porous system from reconstructed X-Ray CT images and to quantify pore-space parameters such as porosity, specific surface area, tortuosity, or network structure, a binarization process is commonly implemented to separate the image into two phases, the solid matrix and pores. In the presented study, we tested a variety of segmentation methods for binarization of CT images of geological and artificial materials. A number of well-established and recently developed techniques ranging from simple image histogram shape and global thresholding to locally adaptive techniques based on geostatistical analysis were implemented and tested. Geometrical and topological pore network properties were determined based on mathematical morphology operations, and then compared not only among different segmentation methods but also to experimentally and analytically obtained values in order to evaluate the effect of these different techniques on quantitative spatial analysis of pore structures.