Computer vision and texture analysis for characterization of X-ray images of rocks

X-ray micro-computed tomography (micro-CT) has provided us digital twins of physical rocks. Digital twins are raw grayscale micro-CT images that contain information regarding rock-microstructure. These raw micro-CT images are then transformed into their segmented counterparts, which attempt to identify different rock materials as labels based on gray-level intensities. However, there are important limitations when using segmented images to understand rock-structure, and direct use of grayscale micro-CT images is warranted. These limitations/challenges include (1) lack of structural information of pore-space geometry during segmentation of grayscale micro-CT images, (2) operator-biased segmentation techniques which cause the rock statistics to be biased, and (3) utilization of grayscale micro-CT images to infer representative elementary volumes (REVs). First, gray-level intensity-based segmentation fails to distinguish pore-spaces with different geometrical characteristics (fractures and granular-pores) because the same gray-level intensities represent them. A combination of computer vision and clustering algorithm is applied to segmented images of fractured rocks to tackle this problem. This workflow allows the segregation of fractures and granular pores into different labels based on their structural properties. Such an analysis adds a structural aspect to segmentation procedures to understand competing pathways for fluid-flow better. Second, a computer vision technique called Gray-Level Co-occurrence Matrix (GLCM) is applied to grayscale micro-CT images. GLCM incorporates directional, spatial, and frequency information of gray-level intensities in a 2D spatial map. The resulting GLCM-statistics are linked to grain-size distributions, (an)isotropic grain arrangement, and mineralogical heterogeneity in Sandstones. Analysis by GLCM alleviates user-biases and allows automation of micro-CT analyses using grayscale statistical measures. Third, a novel texture-characterization technique called Gray-Level Size-Zone Matrix (GLSZM) is used to address the long-standing challenge of inferring grayscale representative elementary volume (GREV) sizes directly from raw micro-CT images. GLSZM-based statistics analogous to porosity and permeability are used to estimate GREVs. GREVs encapsulate details regarding the connectivity of pores and minerals, mineralogical variations, and microporous regions, which present a challenge to the existing segmentation routines. This thesis shows the descriptive power of using computer vision and texture analysis for rock characterization and is a step forward to add grayscale analysis to the Digital Rock Physics workflow.

Grantor dc:publisher

UNSW, Sydney

Year dc:date

2020

Author dc:creator

• Singh, Ankita

• Gray-level co-occurrence matrix
• Porous media
• X-ray micro-computed tomography
• Gray-level size-zone matrix
• Representative elementary volume
• Digital rock physics
• Unsupervised machine learning
• Texture analysis
• Grayscale analysis

Statement dc:rights

• open access
• CC BY-NC-ND 3.0
• free_to_read

Licence

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by-nc-nd/3.0/au/

Language dc:language

EN

Identifier

https://doi.org/10.26190/unsworks/22244

OAI identifier oai:identifier

oai:unsworks.library.unsw.edu.au:1959.4/70495