Answers are grounded in this paper and cite their sources.
Computed tomography (CT) enables non-destructive examination of pore structures, but image-based porosity estimates depend on preprocessing, segmentation, and imaging resolution. This study compares four filtering–thresholding configurations for estimating porosity from 6,840 CT slices representing six construction-material classes: CM0, CM5, CM10, CM20, geopolymer mortar, and ultra-high-performance concrete (UHPC). Median and Gaussian filtering were paired separately with adaptive mean-C and adaptive Gaussian thresholding. Pore-area fractions were calculated from segmented images and compared with available vacuum-pycnometer (PYC) reference values. Three specimens per class were prepared from one batch and tested using PYC, with one specimen subsequently selected for CT scanning. Median filtering with adaptive mean-C thresholding (MB_AMCT) showed promising agreement with the supplied reference values for most investigated classes. In the UHPC comparison figure, all four configurations exceeded the PYC reference value, indicating that unresolved fine pores alone cannot explain the discrepancy. Contributions from noise, local intensity variations, threshold settings, pore misclassification, and physical-reference uncertainty may also contribute, and differences in pore accessibility and spatial coverage mean that CT and PYC cannot be directly equivalent. An additional qualitative comparison looked at candidate pore boundaries in five available CT crops, including UHPC, without independent segmentation ground truth. Agreement in total porosity does not establish accurate pore localization or morphology. Independent CT specimen replication, parameter-sensitivity analysis, and unfiltered comparisons were unavailable. The findings therefore describe performance within the investigated dataset. Further evaluation with documented parameters and independently prepared specimens is required to establish reproducibility, ranking stability, and broader applicability.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: