In present paper we analyze five different thresholding schemes for obtaining the CT scan images of rock to binary images which can be utilized for different property simulation. Riddler''s and Otsu''s methods gave the maximum classification efficiency for the rock samples used. Sensitivity analysis was conducted to assess the impact of change of thresholds on simulated properties like porosity, permeability and formation factor. It showed a consistency in the inter-relation between different properties for different thresholds.
Computational Rock Physics is an upcoming research area in field of rock physics. The 3D digital rocks used in computational rock physics are obtained by 3D-CT scan of the core, which produces a set of 2D intensity images which can be used to reconstruct the 3D volume. The intensity of these images corresponds to X-ray attenuation, which in turn responds to density and atomic number of the material (Ketcham and Carlson, 2001). The process of obtaining 3D binary cube consisting of grains and pores from the set of 2D CT scans involve thresholding and interpolating intensity images to form 3D volume.
Different researchers in varied fields have surveyed a number of thresholding techniques to convert gray-scale intensity images to binary images. Delingette et. al. (1997) evaluated three different thresholding methods for 3D reconstruction of industrial parts from CT scan images while Plaumbo, et. al. (1986) explored different algorithms for document image binarization. Sezgin and Sankur (2004) conducted an exhaustive survey of image thresholding techniques. They chose 40 different image thresholding techniques, categorized them and compared the performance on a set of intensity images.
The present paper investigates different thresholding techniques for CT scan images of the rock. It also explores the sensitivity of the simulated properties as porosity, permeability and formation factor with variation in threshold. The sensitivity analysis is further extended to the inter-relations of these properties.
The present study use samples prepared by Kameda (2005). For these samples, loose sand grains from the beach collected from local (San Mataeo County, California) coastal area, Pomponio Beach (PB), California and Año Nuevo (AN) costal dune (eolian sand) (Figure 1). The loose sand grains were mixed with epoxy, packed and 3D images were digitized by High-resolution X-ray tomography (CTscanning).
Analyzing the thresholds.
Histogram based methods.
The thresholding methods evaluated in present paper can be categorized in two classes:These methods analyze the peaks, valleys and curvature of the histogram. Two methods analyzed in this category were Rosenfeld''s Convex Hull Method (Rosenfeld and Torre, 1983) and Riddler''s iterative thresholding method (Riddle and Calvard, 1978). Rosenfeld method is based on shape properties of the histogram. It analyzes concavity of the histogram which can be obtained as theoretical difference between its convex hull and histogram. The points of maximum curvature form the potential candidates for thresholds. Riddler''s iterative thresholding method is centered on the assumption that bimodal histogram is sum of two Gaussian models. At the nth iteration, a new threshold is established by computing the average of the mean of two Gaussian classes (foreground and background) obtained by threshold at the previous step.