Effect of scanning and image reconstruction settings in X-ray computed microtomography on quality and segmentation of 3D soil images

Over the last decade, X-ray computed tomography (CT) has been used increasingly to characterise the microscale architecture of soils. As a result significant progress has been made in the acquisition and interpretation of X-ray CT data, as well as in the thresholding of 3D greyscale CT images in order to produce binary (black and white) ones. Nevertheless, sizeable uncertainties persist, in particular concerning optimal instrumental settings used to generate the greyscale images. In this context, the key aim of this study is to investigate in detail the effect of scanning resolution and reconstruction settings such as noise reduction and 32-bit to 8-bit mapping interval on the 3D X-ray CT imaging of soil structure and the impact on the performance of thresholding methods. To assess the quality of the X-ray CT greyscale images, measures of contrast, noise and sharpness are proposed and tested on a series of images of five different soil samples. At the same time, performance of four segmentation algorithms, i.e., three methods recently developed to deal specifically with soil samples and Otsu's method as a benchmark, was evaluated using functional measures of 3D binary images, including Minkowski functionals and surface pore connected fraction. Results of these analyses suggest that the acquisition and reconstruction parameters investigated significantly affect the quality of soil images, and the subsequent thresholding process. In particular, it was found that thresholding by any of the four methods is greatly affected by the quality of image sharpness, which for soil images appears to be mainly controlled by the scanning resolution. As a result, it is concluded that no matter what reconstruction resolution is required in a study, in order to allow an accurate identification of the pore space, the sample should always be scanned at the highest resolution permitted by the scanning instrument and the sample size. Results also suggest that the three segmentation methods recently developed for soil images thresholding are robust to different levels of noise as well as the choice of the 32-bit mapping interval, as long as lower and upper interval limits for mapping are chosen within suitable boundaries.