On the challenges of greyscale‐based quantifications using X‐ray computed microtomography

Abstract

For X-ray computed microtomography (μ-CT) images of porous rocks where the grains and pores are not fully resolved, the greyscale values of each voxel can be used for quantitative calculations. This study addresses the challenges that arise with greyscale-based quantifications by conducting experiments designed to investigate the sources of error/uncertainty. We conduct greyscale-based calculations of porosity, concentration and diffusivity from various μ-CT experiments using a Bentheimer sandstone sample. The dry sandstone is imaged overtime to test the variation of greyscale values over sequential scans due to instrumentation stability. The sandstone is then imaged in a dry and contrastagent saturated state at low resolution to determine a porosity map, which is compared to a porosity map derived from segmented high-resolution data. Then the linearity of the relationship between the concentration of a contrast agent and its corresponding attenuation coefficient is tested by imaging various solutions of known concentration. Lastly, a diffusion experiment is imaged at low resolution under dynamic conditions to determine local diffusivity values for the sandstone, which is compared to values derived from direct porescale simulations using high-resolution data. Overall, we identify the main errors associated with greyscale-based quantification and provide practical suggestions to alleviate these issues.

A New Upscaling Method for Fluid Flow Simulation in Highly Heterogeneous Unconventional Reservoirs

“…The core structure can be divided into pores and matrix by the binary method through the gray map. Moreover, there are much other information that can be obtained from grayscale CT images, such as local porosity and mass concentration of the transported component ( [29]):…”

Review of Data Science Trends and Issues in Porous Media Research With a Focus on Image‐Based Techniques

“…Detection of void and solid spaces is not an easy procedure in the most of porous material images (Zhang et al, 2019). Gray-level of the images can change between 0 and 1, and determining a certain thresholding level is not viable in many cases due to several reasons including but not limited to:…”

“…Detection of void and solid spaces is not an easy procedure in the most of porous material images (Zhang et al., 2019). Gray‐level of the images can change between 0 and 1, and determining a certain thresholding level is not viable in many cases due to several reasons including but not limited to: Low spatial resolution of the images in the presence of micro‐porosity that creates pixels with intensities at the middle of the range (Bultreys et al., 2015) Shadows and artifacts caused by change in the wave speed at the inter‐phase boundary that can lead to overestimate or underestimate the porosity Gradient in the average intensity that prevent using methods, such as global Otsu thresholding (Otsu, 1979) to separate solid and void spaces (Freire‐Gormaly et al., 2015; Iassonov et al., 2009) Reconstruction noises due to the limited projection angles that will add radial artifact shadows to the image known as streaking patterns (Han et al., 2016) …”

DeePore: a deep learning workflow for rapid and comprehensive characterization of porous materials

“…Alqahtani et al [46] used CNNs to estimate porosity, average pore size and specific surface of the porous rocks based on both types of 2-D tomography images and found that binary images could give a more accurate estimation of porous material characteristics compared to the gray-scale ones. However, the morphology of the binarized images is highly dependent on the thresholding technique and it suffers from an inherent uncertainty [63]. In another attempt, Cang et al [64] designed a CNN for prediction of physical properties of heterogeneous materials and successfully predicted the Young modulus, diffusion and permeability of the porous material with more than 90% of certainty on their testing data.…”