X-ray computed tomography (CT) has become increasingly popular for non-destructive investigation of cementitious systems. Differentiating solid phases in the reconstructed image volume has been reported as a challenge due to similar greyscale values, thus limiting x-ray CT to qualitative or semi-quantitative analysis for solid phases. The goal of this research was to investigate image preprocessing and segmentation algorithms in quantifying synthesized hydration products. C-S-H, calcium hydroxide, monosulfate, and ettringite were created in the laboratory for investigation. In this research, two local segmentation algorithms, Markov random fields and watershed, and one global segmentation algorithm, Gaussian histogram deconstruction, were studied for their ability to quantify the phases. It was found that misclassification of voxels occurred when using the Gaussian deconstruction. An average reduction of 50%, and as high as 180%, in the quantification errors were observed when using local segmentation algorithms.

Tyler Deboodt, Dorthe Wildenschild, Jason H. Ideker, O. Burkan Isgor, Comparison of thresholding techniques for quantifying portland cement hydrates using synchrotron microtomography, Construction and Building Materials, Volume 266, Part B, 2021, 121109, abstract

Comparison of segmentation algorithms in a quaternary mixture of phases from sample Q2-1. Blue represents AFt, yellow represents AFm, purple represents C-S-H, and green represents CH. a) Greyscale orthoslice of a quaternary mixture. b) Segmented using Gaussian deconstruction. c) Segmented using watershed. d) Segmented using MRF. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)