Abstract:
This thesis investigates the problem of segmenting X-ray computed tomography (CT) scans of soil samples by thresholding image signals (intensities, or grey levels). Comparisons of different techniques have shown that the two-dimensional (2D) binomial locally adaptive indicator kriging (LAIK) provides most robust separation of pore and solid matters. The LAIK adapts the grey value of each ambiguous image point on the basis of the neighbouring grey levels (with due account of their geometric closeness) and the imagewide grey level probability distribution. The thesis modifies and optimises the LAIK algorithm by extending it to the 3D CT scans and allowing for various numbers of classes (objects) at each 2D section of the 3D image. Also, the thesis proposes, implements, and experimentally investigates a novel standalone Java-based automated segmentation tool to facilitate comparing and improving the soil CT segmentation methods.