Combining global and local scaling methods to detect soil pore space

Abstract

Computed tomography (CT) images provide very useful information to characterize the pore space structure in different porous materials. Being a non-destructive technique, it allows preserving the real characteristics of the analysed samples. Moreover, providing 3 dimensional (3D) information, it also allows making more exhaustive studies on other characteristics of pore space such as connectivity or pore geometry. However, this technique also has drawbacks such as the insertion of errors and distortions from CT reconstruction that could drive to low contrast greyscale images hindering the detection of solid-void interface.The Singularity-CA (S-CA) binarization method has already proven to be an efficient method when detecting the pore space in CT soil images. This method is primarily based on the existence of self-similar properties in the singularity value (SV) spatial distribution. These self-similar properties derive to thresholds in the singularity map which are used to delimit the pore space. This method detects the medium and large pore sizes with an excellent fit. However, due to its high sensibility to low fluctuations in the greyscale images, some small pores are incorrectly detected and therefore its amount is overestimated.This work attempts to solve this drawback introducing a new approach named the Combining Singularity-CA (CS-CA) method. This improved method is based on the combination of a global thresholding method, the Maximum Entropy method, and the S-CA method. The CS-CA methodology is fully automatic and did not require human adjustment. A set of soil synthetic images with porosities from 3% to 25% were used to validate the new approach. These images were constructed using the Truncated Multifractal (TM) method, especially useful to simulate CT soil images. Based on parameters such as porosity, relative error and misclassification error, CS-CA method gave better pore detection than the S-CA and the Maximum Entropy method applied individually to the images.Finally, it was compared the two main methods used to detect slope-change points in plots with several linear segments: the linear regression (LR) method and the wavelet transform modulus maxima (WTMM) method. These methods are very important in the S-CA methodology since they are responsible of defining the threshold in the binarization step. The WTMM method proved to be easier to deal with when using by the new CS-CA method.