Application of Segmentation for Correction of Intensity Bias in X-Ray Computed Tomography Images

Abstract

Nondestructive imaging methods such as x‐ray computed tomography (CT) yield high‐resolution, grayscale, three‐dimensional visualizations of pore structures and fluid interfaces in porous media. To separate solid and fluid phases for quantitative analysis and fluid dynamics modeling, segmentation is applied to convert grayscale CT volumes to discrete representations of media pore space. Unfortunately, x‐ray CT is not free of artifacts, which complicates segmentation and quantitative image analysis due to obscuration of significant features or misinterpretation of attenuation values of a single material in different image sections. Images or volumes emanating from polychromatic (industrial) scanners are especially prone to high noise levels, beam hardening, scattered x‐rays, or ring artifacts. These problems can be alleviated to a certain extent through application of metal filters, careful detector calibration, and sample centering, but they cannot be completely avoided. We have developed a simple three‐dimensional approach to numerically correct for image artifacts using sequential segmentation. This procedure leads to a significant improvement of grayscale data as well as final segmentation results with reasonable computational demand.

Segmentation of X‐ray computed tomography images of porous materials: A crucial step for characterization and quantitative analysis of pore structures

“…Another problem we encountered when using the FLASHCT TM system was the distortion of attenuation intensities during CT reconstruction, resulting in different apparent gray scale intensities in the reconstructed CT image corresponding to the same material at different heights in the scanned sample column. Since such distortions pose an insurmountable problem for all global thresholding techniques, we have developed a correction procedure to account for large-scale smooth intensity variations in 3-D CT images [Iassonov and Tuller, 2009]. Application of this correction is illustrated in Figure 2.…”

Quantifying fracture geometry with X-ray tomography: Technique of Iterative Local Thresholding (TILT) for 3D image segmentation

“…The gray box and arrows represent the step for the generation of the fracture template. The detailed procedures of the MHF filter are provided in the workflow in the supplementary materials thresholding method which minimizes intra-class variations and maximizes inter-class variations, and has been demonstrated to be a computationally efficient and robust algorithm [59,68,69]. The resulting coarsely segmented image then undergoes post-processing to distinguish the fracture from pores.…”

X-ray microtomography analysis of soil structure deformation caused by centrifugation

“…This version of the non-local means filter requires two parameters: (1) the radius of the search window is set to four pixels and an estimate of the noise level expressed as a standard deviation of noise is set to 60; (2) vertical differences in average image intensity due to shading and cone beam artifacts are removed. To do so, the mean gray value of the soil matrix in a specific z-slice is measured and the difference to the mean gray value of the soil matrix in the entire image is subtracted from each voxel in that depth (Iassonov and Tuller, 2010). The two thresholds that separate the soil matrix from darker pore voxels and brighter rock voxels are chosen manually and do not effect the results much as long as they cover the entire grayscale range of soil without adding pores and rocks to the average.…”