Assessing the Impact of Pre-Segmentation Image Processing Filters on the Accuracy of Rock Characteristics Derived from CT Datasets

Abstract:

Accurate representation of pore structures in complex subsurface rock formations is essential to improving our understanding of transport and fate of fluids in the subsurface. X-ray computed tomography (CT) allows nondestructive 3D imaging of the pore space, but segmentation of CT datasets into pore and non-pore space is nontrivial. Improving the precision and accuracy of this step is crucial because core-scale porosity is used to infer upscaled reservoir parameters including permeability and reactive mineral surface areas. In this study, an Indiana Limestone core was characterized in two CT scans, taken before and after exposure to acidic fluid flow. A series of image processing filters was applied to the CT datasets in preparation for training class definition using the machine learning-based Trainable Weka Segmentation (TWS) plugin in Fiji. These filters include an unsharp mask to enhance feature boundaries, a bilateral filter to reduce noise, and a beam-hardening correction to correct for CT imaging artifacts. The variability of output porosity with and without these filters was assessed in this study. The application of all filters in combination was associated with a reduction in porosity variance, though all porosity values found using this method fell below experimentally determined porosity. This is likely due to the presence of sub-voxel porosity in the relatively low-resolution CT datasets (voxel resolution 28 μm). Pore size distribution (PSD) was experimentally determined using mercury intrusion porosimetry (MIP), and MIP results were compared to PSD results derived from CT segmentation. 3D computational meshes of the segmented datasets were used to simulate flow in the cores using OpenFOAM to estimate permeability. Simulated permeability values were compared to laboratory experimental values. The results of this study emphasize the importance of careful consideration of image processing techniques applied to CT imagery for digital rock characterization.