Professional Documents
Culture Documents
A R T I C L E I N F O A B S T R A C T
Keywords: An image processing workflow is presented for the characterization of pore and grain size distributions in porous
X-ray microcomputed tomography (μCT) geological samples from X-ray microcomputed tomography (μCT) and scanning electron microscopy (SEM)
Scanning electron microscopy (SEM) images. The pore and grain size distributions of five sandstone samples including Berea, Buff Berea, Nugget,
Pore size distribution
Castlegate, and Bentheimer, and one carbonate sample, Indiana limestone, are extracted using the proposed
Grain size distribution
Image processing
workflow. Two-dimensional size distributions acquired from SEM images were found to be biased toward smaller
Rocks sizes misrepresenting the actual 3D distributions. Stereological techniques unfolded the measured 2D size dis
Porous materials tributions from SEM images to 3D distributions comparable with μCT results. While larger pores and grains can
easily be detected from μCT and SEM images, the quantification of small-scale heterogeneities is severely
influenced by their limits of resolution. We show that microstructural details resolved by SEM can significantly
impact the pore and grain size distributions in sandstone and carbonate rock samples. For example, SEM-resolved
microporosities in Indiana limestone result in bimodal distributions of pore and grain sizes, whereas μCT ob
servations exhibit unimodal distributions. The acquired images and processed results are openly available and
may be used by researchers investigating image processing, magnetic resonance relaxation or fluid flow simu
lations in natural rocks. The proposed methodology can be implemented to process μCT and SEM images of
natural rocks as well as other types of porous materials.
* Corresponding author.
E-mail address: armafr@dtu.dk (A. Afrough).
https://doi.org/10.1016/j.cageo.2021.104895
Received 23 March 2021; Received in revised form 28 June 2021; Accepted 20 July 2021
Available online 26 July 2021
0098-3004/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
H. Safari et al. Computers and Geosciences 156 (2021) 104895
production in wellbores drilled in unconsolidated and consolidated In this work, we propose an image processing workflow for
sandstone rocks (Ghasemi et al., 2018; Higgins, 2000; Srisutthiyakorn extracting pore and grain size distributions of natural rocks from X-ray
and Mavko, 2019). microcomputed tomography and scanning electron microscopy images.
Various experimental methods have been developed for analyzing We employ stereological techniques to convert 2D distributions ob
the pore and grain size distributions in porous rocks over the years. tained from SEM images to 3D distributions comparable with μCT re
Practical methods for determining the grain size distribution include sults. While most commercial image processing software do not offer
sieve analysis, laser diffraction, point-count method, and more recently stereological algorithms, this study emphasizes the role of stereology in
the dynamic image analysis method (Roostaei et al., 2020; Srisutthiya obtaining representative pore and grain size distributions from 2D SEM
korn and Mavko, 2019). Although these methods are widely used for images. Pore and grain size distributions are presented in terms of
determining the particle size and shape in unconsolidated materials, volumetric probabilities as opposed to frequency- and count-based
they cannot be applied to consolidated porous media (Ghasemi et al., methods. Volume fractions are of significant importance in engineer
2018). Experimental methods for measuring the pore size distribution ing and theoretical calculations concerning aquifer capacity, distribu
include mercury injection porosimetry (MIP) for the pore throat size tion of fluid saturations in oil reservoirs, and geo-mechanical properties
distribution, nitrogen adsorption isotherms for the pore body size dis of rocks. The pore and grain size distribution of six common
tribution of micropores, and nuclear magnetic resonance (NMR) for the commercially-accessible rock samples, including five sandstones and
pore body size distribution (Afrough et al., 2019; Arns, 2004; Kruschwitz one carbonate, are investigated. The rock samples span a variety of rock
et al., 2020; Shuaibing Song et al., 2019). These methods normally types with different and similar pore and grain size distributions and are
require the injection of a fluid phase into porous samples and mostly used by researchers worldwide in their verifying of measurement
describe the spatial distribution of connected pore space. As a result, methods, such as in magnetic resonance relaxation (Afrough et al., 2019;
disconnected pore structures which might become reconnected in pro Arns, 2004). The algorithms presented in this study could directly be
cesses such as reactive fluid transport still remain undetected (Al-Khu applied or easily extended to other porous materials. The acquired im
laifi, 2018). ages and obtained results can also be used by those processing μCT and
The emergence of X-ray microcomputed tomography (μCT) as a non- SEM images, analyzing magnetic resonance relaxation, or those per
invasive and non-destructive method along with recent improvements in forming fluid flow simulation in rocks.
the computational power has made it possible to visualize the internal
structure of porous materials. The μCT scanning techniques have been 2. Material and methods
broadly used to characterize the geometrical structure of pore, grain,
and mineral phases in rock samples (Arns, 2004; Elkhoury et al., 2019; 2.1. Description of rock samples
Ghasemi et al., 2018; Lai et al., 2015; Schmitt et al., 2015; Srisutthiya
korn and Mavko, 2019; Zhang et al., 2018). The resolution of 3D scans is Five different sandstone samples including Berea (BA), Nugget (NU),
directly proportional to the sample size, beam quality, and specifications Buff Berea (BB), Castlegate (CG), and Bentheimer (BT), and one car
of the detector (Blunt et al., 2013). The true resolution of μCT systems is bonate sample, Indiana Limestone (IL), were acquired from Kocurek
limited to approximately 0.3 μm in projection imaging, in comparison Industries (Caldwell, TX, USA). The selected rock samples present a wide
with a sub-50 nm imaging resolution in lens-based X-ray CT systems range of pore and grain size distributions.
(Withers, 2007). High-resolution scans can reveal submicron textures Berea and Buff Berea samples are Upper Devonian grain-supported
and heterogeneities in porous materials at the cost of significant sample sandstones from the Kipton formation. Porosity of the Berea sample
size reduction (Bai et al., 2013; Blunt et al., 2013). Scanning electron ranges from 18 to 21 percent, while Buff Berea has slightly higher po
microscopy (SEM) is another method widely employed in describing the rosities ranging from 20 to 22 percent. Castlegate is a Late Cretaceous
submicron structure of porous materials. SEM offers high-resolution sandstone sample from the Price River formation of Mesaverde group
images with limited field of views (FOVs) that could be used in the which has a porosity ranging from 26 to 29 percent. Nugget is a Late
characterization of small-scale heterogeneities in natural rocks. SEM, in Triassic heterogeneous sandstone sample from the Utah formation
the backscattered electron detection mode, can observe compositional which compared to other samples has a lower porosity ranging from 10
variations with a lateral spatial resolution of approximately 10–100 nm to 15 percent. Bentheimer is a well-sorted Valaginian sandstone from the
depending on the specimen composition and the beam energy selected Germany formation and has a porosity ranging between 21 and 26
(Goldstein et al., 2018). Two-dimensional SEM images have been mostly percent. The sandstone samples are composed dominantly of quartz,
considered as a qualitative tool and their quantitative analysis to obtain with other present minerals being feldspar, clays, and sometimes calcite
pore and grain size distributions have been limited (Song et al., 2019; S. (Afrough et al., 2017; Dalton et al., 2020).
Song et al., 2019b; Zhang et al., 2015). Indiana limestone is a Mississippian fossiliferous limestone from the
Two-dimensional size distributions obtained from SEM images do Salem limestone formation in the Bedford-Bloomington area which is
not represent the actual 3D distributions because of the sectioning bias mainly composed of calcite (more than 97 %) with minor amounts of
(Liu et al., 2020; Reppel and Weinberg, 2019; Song et al., 2019b). Sri other minerals, such as alumina, iron oxide, silica, and magnesite pre
sutthiyakorn and Mavko (2019) presented a general approach for con sent (Ji et al., 2012). Its porosity with a bimodal pore size distribution is
verting 2D grain size distributions acquired from thin section images to characterized by intergranular micro-connected macropores and ranges
3D distributions by providing a forward solution to the Wicksell’s between 12 and 19 percent.
corpuscle problem. They obtained transformation matrices relating 2D
(from CT slices) to 3D grain size distributions using μCT images of 2.2. X-ray microcomputed tomography
natural rocks. We show in this study how their method could fail when
inherent microstructures in porous rocks are not properly resolved in Standard 1.5 inch core plugs were cut into 2–3 mm thick slabs. Small
μCT images. A variety of methods have been presented in the literature samples were chipped from these slabs and installed on stubs by glue.
for the characterization of pore and grain size distribution in porous These samples had varying dimensions approximately close to 2 × 2 × 4
geological samples. However, proposed methodologies are limited from mm3, but were not perfectly shaped. High-resolution μCT images were
one image processing environment to another and are not checked acquired using a Skyscan 1072 system at the Microscopy and Micro
against several other methods and several samples. Therefore, a general analysis Facility of the University of New Brunswick in Fredericton,
image processing workflow that could be applied to both 2D and 3D Canada. Some of the samples were slightly larger than the field of view
images of rock samples for quantitative analysis of pore and grain size and were imaged in the region of interest imaging mode. Imaging was
distributions is still lacking. performed with a tube voltage of 91 kV and a source current of 110 μA
2
H. Safari et al. Computers and Geosciences 156 (2021) 104895
The acquired 16-bit images were first converted into 8-bit images
and their beam hardening and region of interest imaging artifacts were
corrected. The maximum possible rectangular sub-volumes were
extracted from each image for the quantification of the samples’ pore
and grain size distributions (see Appendix). Fig. 1 shows the general
workflow employed in this study for analyzing the μCT images. To
improve the signal-to-noise ratio, a 2D edge-preserving non-local means
filter was applied on every slice in the xy-plane. Two-dimensional,
instead of 3D, non-local means filtering was employed because of the
poor computational performance of the 3D filter in Avizo, even in the Fig. 1. The general workflow employed in the extraction of pore and grain size
GPU-processing mode. Table 2 summarizes the non-local means filter distributions from μCT images. An edge-preserving non-local means filter is
applied for denoising purposes and the filtered images are segmented into two
settings along with other image parameters including image resolution,
binary images representing pores and grains. A Euclidean distance transform is
and dimensions of the extracted sub-volumes. The pore size distribution
applied on the binary images which is then masked by a medial axis transform.
was not very sensitive to the variation of the search window size in the An H-maxima filter is applied on the masked distance map to obtain markers for
non-local means filtering. watershed segmentation. The labeled images from the watershed algorithm are
In the next step, we applied image segmentation techniques to used to obtain volumetric pore and grain size distributions. Each side of a cube
segment each gray-scale image into two binary images representing the is 675 μm.
pores and grains of the rock samples. This is the most crucial step as it
affects all the subsequent quantitative analysis. Although a variety of transform was applied to the segmented images of pores and grains. A
segmentation algorithms have been proposed in the literature, global medial axis transform was applied to binary images to extract connected
thresholding technique is still the most commonly applied approach in voxel skeletons which was then used to mask the calculated distance
porous media research (Iassonov et al., 2009). Automatic thresholding maps. An H-Maxima filter was applied on the new distance map to find
algorithms such as Otsu’s method (Otsu, 1979) tended to over-segment the peaks in every region. The calculated peaks were passed as seeds to a
the gray-scale images in almost all the samples. This could be attributed marker-based watershed segmentation algorithm to provide labeled
to their sensitivity to partial volume effects arising from the unresolved images for pore and grain volumes. Segmented regions were analyzed
sub-resolution features and limited contrast-to-noise ratio in μCT images and equivalent diameters, defined as the diameter of a sphere of the
(Sheppard et al., 2014). An interactive thresholding procedure with same volume as that of a segmented region, were calculated. Therefore,
immediate visual feedback was found to provide more accurate and the term diameter in the context of pore and grain diameters is referring
consistent segmentation results. However, manual thresholding still to the equivalent diameter throughout this paper. These spheres with
remains labor intensive and might introduce some operator bias. varying equivalent diameters are represented as nodes in the generated
Having performed image segmentation, a Euclidean distance pore and grain network models. Volumetric probability functions for
pore and grain size distributions were calculated using the regional
Table 1 volumes and their respective equivalent diameters.
Typical resolution scales of the captured SEM images for each sample.
Samples NU CG BT BB BA IL
3
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Table 2
Parameters of the μCT images including their resolution, the dimensions of extracted sub-volumes, and the non-local means filter settings for each sample.
Parameters NU CG BT BB BA IL
4
H. Safari et al. Computers and Geosciences 156 (2021) 104895
process.
In Fig. 3, the occurrence frequency ϕ, for spherical particles with where i = 1, .., n with n being the number of bins in the histogram, Hi is
′
radius R intersecting with circles of radius r is: the mean projected height which is equal to 1 for spherical particles, α1
is the conversion coefficient of the first class, and P represents the cross-
section size probabilities. The probability of a particular class in the
cross section with radius r (r1 < r < r2 ) intersecting with a sphere of
radius R is given by:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
1
P(r|R) = ( (R2 − r21 ) − (R2 − r22 )) (4)
R
where r1 and r2 are the lower and upper limits of the classes in 2D his
togram, respectively. Mathematical details of the above formulations
could be found in the work of Sahagian and Proussevitch (1998).
The “GrainSize Tools” Python scripts developed by Lopez-Sanchez
(2018) for grain size analysis were modified to account for the volu
metric probabilities and employed in this study to calculate 3D distri
butions of pore and grain sizes from 2D distributions.
5
H. Safari et al. Computers and Geosciences 156 (2021) 104895
6
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Fig. 5. Typical SEM images of sandstone and carbonate samples and their multiphase segmentation results. Sandstone samples are mainly composed of quartz grains
and small amounts of feldspar and clays. Clay minerals are observed either as thin layers or as agglomerates filling the pore space. Indiana limestone features a
monomineralic texture predominantly composed of calcite mineral. Well-resolved SEM images reveal microporosity regions in sandstones and Indiana limestone. The
scale bars are equivalent to 500 μm except for Berea sandstone which has a scale bar equivalent to 200 μm.
Table 3 Table 4
Comparison of the experimental porosities obtained from gravimetry with those Area fractions of the minerals present in sandstone samples derived from SEM
calculated from μCT and SEM images. observations.
Porosity (%) NU CG BT BB BA IL Minerals NU CG BT BB BA
7
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Fig. 6. The volumetric probability of pore diameters from μCT (blue) and unfolded SEM (red) for Bentheimer (BT), Berea (BA), and Indiana limestone (IL) (Top), and
for Castlegate (CG), Buff Berea (BB), and Nugget (NU) (Bottom). SEM-resolved microporosity regions in clay minerals has shifted pore size distributions of sandstone
samples toward smaller sizes. The bimodal pore size distribution obtained from SEM images of Indiana limestone is also attributed to highly resolved microporosity
regions at the grain boundaries. The pore size distributions from μCT and SEM images are consistent. (For interpretation of the references to colour in this figure
legend, the reader is referred to the Web version of this article.)
Table 5
The statistical description of pore diameters from μCT, 2D SEM, and the unfolded 3D SEM distributions for all rock samples. All pore diameters are in μm.
Statistical Parameters NU IL CG
μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D)
Statistical Parameters BT BB BA
μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D)
pore space. These agglomerates even resulted in a bimodal grain size size distribution is observed for feldspar. The calcite mineral in Indiana
distribution of clays in Berea sandstone (see Fig. 7). Quartz and feldspar limestone also exhibits a bimodal grain size distribution consistent with
minerals show similar distributions with grain sizes spanning from 1 to its highly resolved microstructural details in SEM images.
250 μm. These minerals demonstrate a unimodal grain size distribution To compare the grain size distribution histograms with those ob
in all sandstone samples, except for Castlegate where a bimodal grain tained from μCT, the mineral phases were merged into a single solid
8
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Fig. 7. The volumetric probability of mineral grain diameters from SEM images for Bentheimer (BT), Berea (BA), and Indiana limestone(IL) (Top), and for Castlegate
(CG), Buff Berea (BB), and Nugget (NU) (Bottom). Clay minerals in sandstones exhibit a wide range of grain sizes ranging from 1 to 40 μm. The appearance of large
grain sizes are due to agglomeration of clays within the pore space resulting in a bimodal grain size distribution in Berea sandstone. The distribution of grain sizes in
quartz and feldspar minerals is unimodal in almost all sandstone samples except for Castlegate sandstone where a bimodal grain size distribution is observed for
Feldspar. Calcite in Indiana limestone also shows a bimodal grain size distribution due to its well-resolved microstructural details from SEM images.
Table 6
Statistical description of the quartz, feldspar, and clay grain diameters from the unfolded 3D SEM distributions for sandstone samples. All grain diameters are in μm.
Statistical Parameters NU CG BT
Statistical Parameters BB BA
Quartz Feldspar Clay Quartz Feldspar Clay
mean 96 61 12 44 38 9.6
std 47 51 7.8 35 35 9.5
min 39 11 3.3 1.0 1.1 1.0
25 % 57 22 6.1 17 10 3.3
50 % 85 43 10 34 24 6.4
75 % 127 86 18 65 55 13
max 188 190 30 125 127 46
kurtosis 0.7 0.6 0.5 0.2 0.3 5.6
skewness 0.6 1.2 0.8 0.9 1.1 2.1
phase. Fig. 8 illustrates the volumetric probabilities of grain size dis agreement between volume-weighted grain sizes, the merging of min
tribution in sandstone and carbonate rock samples. Although the 3D eral grain components resulted in large arithmetic mean grain diameters
volumetric grain size distributions obtained from SEM images are from SEM images, compared to those of μCT results. Similar to pore size
largely consistent with μCT results, the probabilities of small-scale distributions, 2D grain size distributions from SEM images represent
grains like those of clay minerals have vanished due to merging. high values of kurtosis and skewness compared to 3D distributions.
Table 7 provides the statistical description of the grain diameters Among sandstones, Bentheimer has the largest average grain diam
derived from μCT and SEM images for all samples. Despite the eter (82 μm from μCT and 208 μm from SEM), while Berea represents the
9
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Fig. 8. Volumetric probability of grain diameters from μCT (blue) and unfolded SEM (red) for Bentheimer (BT), Berea (BA), and Indiana limestone (IL) (Top), and for
Castlegate (CG), Buff Berea (BB), and Nugget (NU) (Bottom). The merging process resulted in the disappearance of small-scale grains like clays in SEM images and
shifted the mean grain diameters toward larger values. Despite discrepancies in arithmetic means, volume-weighted averages of grain diameters from SEM images are
largely consistent with μCT results. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
Table 7
Statistical description of the grain diameters from μCT, 2D SEM, and the unfolded 3D SEM distributions for all rock samples. All grain diameters are in μm.
Statistical Parameters NU IL CG
μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D)
Statistical Parameters BT BB BA
μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D) μCT SEM (2D) SEM (3D)
smallest (51 μm from μCT and 95 μm from SEM). Castlegate, Buff Berea, 4.3. Representativeness
and Nugget represent relatively similar mean grain diameters. For
Indiana limestone, the average grain diameter from μCT is 71, whereas All the rock samples used in this study showed homogeneous char
the mean grain sizes from SEM observations are 12 and 88 for small and acteristics. The agreements between the pore and grain size distributions
large grains, respectively. The grain size distribution of Indiana lime obtained from μCT and SEM images confirm this homogeneity and
stone obtained from μCT exhibits a left-tailed unimodal distribution indicate that the selected volumes for μCT imaging are fairly represen
skewed toward smaller grain diameters. tative. Yet, discrepancies exist between the amount of microstructural
details, such as small pores and grains, resolved in μCT and SEM images
10
H. Safari et al. Computers and Geosciences 156 (2021) 104895
which is attributed to their limits of resolution. High resolution μCT minerals within the pore space resulted in a bimodal grain size distri
imaging requires even smaller sample sizes which might result in their bution in Berea sandstone. The clay microporosities were also observed
loss of representativeness as compared to other characterization meth to influence the pore size distributions in sandstone samples. In Indiana
odologies. For the rock samples considered in this study, the pore vol limestone, the SEM-resolved microporosity regions led to bimodal pore
ume fractions from μCT images were quite consistent with experimental and grain size distributions as opposed to unimodal distributions ob
porosities obtained by gravimetry, whereas void fractions derived from tained from μCT images.
SEM observations presented noticeable discrepancies. The mean pore We showed that our workflow can be successfully employed to
sizes obtained from μCT and SEM images, however, were found to be in process and extract quantitative information from μCT and SEM images
close agreement with each other as well as with the absolute pore sizes of natural rocks. We addressed the impact of sectioning bias in SEM
measured using magnetic resonance relaxation. Unlike μCT and SEM images and highlighted the role of stereology in obtaining representa
imaging techniques, experimental methods such as NMR and MIP, only tive 3D size distributions. The rock materials implemented in this study
provide information about the connected pore space. span a variety of rock types with different and similar pore and grain size
distributions. These rock samples are widely used by researchers
5. Conclusions worldwide for analyzing magnetic resonance relaxation or performing
pore-scale fluid flow simulations.
We have shown the application of a proposed image processing
workflow for the extraction of pore and grain size distributions from X-
ray microcomputed tomography and scanning electron microscopy im Declaration of competing interest
ages in five different sandstones and one carbonate sample. While it was
difficult to distinguish between the mineral phases in rock samples from The authors declare that they have no known competing financial
μCT images, particularly clay minerals in sandstones, SEM images pro interests or personal relationships that could have appeared to influence
vided a sufficient contrast to observe different minerals from 2D cross- the work reported in this paper.
sections. The predominant minerals in sandstone rocks were quartz,
feldspar, and clays, whereas Indiana limestone exhibited a mono Acknowledgements
mineralic texture of calcite mineral. Two-dimensional pore and grain
size distributions from SEM images were found to be misrepresenting Authors thank Steven R. Cogswell from Microscopy and Microanal
the actual 3D distributions due to the sectioning bias. Stereological ysis for microscopy, Stephen Delahunty from the Earth Sciences
methods converted the 2D size distributions obtained from SEM images Department for sample preparation, Tatiana Zaraiskaya from UNB Li
to 3D distributions and the results were compared with μCT observa braries for data management, and Florea Marica from UNB MRI
tions. The distribution histograms derived from processing μCT and SEM Research Center for assistance, all from the University of New Bruns
images presented noticeable differences with regard to their limits of wick. Armin Afrough acknowledges the Danish Hydrocarbon Research
resolution. We observed that microstructural details resolved in SEM and Technology Centre for funding. Bruce J. Balcom acknowledges
images can significantly affect the distribution of pore and grain sizes in NSERC of Canada for a Discovery Grant and the Canada Chairs Program
both sandstone and carbonate samples. The agglomeration of clay for a Chair in Material Science MRI.
Appendix
Figure A.1 illustrates an example of the artifact correction, sub-volume extraction, and nonlocal means filtering for X-ray CT images.
11
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Fig. A.1. Illustration of the image processing steps performed on μCT images, (a) original image (b) artifact correction. There are two different artifacts in the
original image including (1) the region of interest imaging artifact and (2) the beam hardening artifact. We applied the beam hardening module in Avizo software to
correct for the intensity variations caused by both artifacts, (c) maximum possible rectangular sub-volume extraction from the image, (d) cropping of the selected
rectangular sub-volume, and (e) denoising the extracted sub-volume via applying a nonlocal means filter.
Author contributions
The data and code/procedures of this manuscript are available from a data repository.
Data, Python codes, and Avizo procedures associated with this manuscript are available in the UNB institutional repository Dataverse at DOI htt
ps://doi.org/10.25545/A19OPN. The Python code is available as a Jupyter Notebook SEM_Analyzer.ipynb, version 1.0, and the Avizo 2019.1 pro
cedure, including all parameters for all samples, is available in Avizo_Worflow.csv. The data and codes are available as CC0 – Public Domain
Dedication.
12
H. Safari et al. Computers and Geosciences 156 (2021) 104895
13
H. Safari et al. Computers and Geosciences 156 (2021) 104895
Wicksell, S.D., 1925. The corpuscle problem. A mathematical study of a biometric Zhang, Y., Jin, S., Wang, Y., Wang, Y., 2015. Characterization of the pore size
problem. Biometrika 17, 84–99. https://doi.org/10.1093/biomet/17.1-2.84. distribution with SEM images processing for the tight rock. In: 2015 IEEE
Withers, P.J., 2007. X-ray nanotomography. Mater. Today 10, 26–34. https://doi.org/ International Conference on Information and Automation, pp. 653–656. https://doi.
10.1016/S1369-7021(07)70305-X. org/10.1109/ICInfA.2015.7279367.
Yang, Y., Liu, P., Zhang, W., Liu, Z., Sun, H., Zhang, L., Zhao, J., Song, W., Liu, L., An, S., Zhang, Z., Kruschwitz, S., Weller, A., Halisch, M., 2018. Enhanced pore space analysis by
Yao, J., 2016. Effect of the pore size distribution on the displacement efficiency of use of μ-CT, MIP, NMR, and SIP. Solid Earth 9, 1225–1238. https://doi.org/
multiphase flow in porous media. Open Phys. 14, 610–616. https://doi.org/ 10.5194/se-9-1225-2018.
10.1515/phys-2016-0069.
14