Segmentation of irregular porous particles of various sizes from X-ray microfocus computer tomography images using a novel adaptive watershed approach

Abstract

This paper presents a novel adaptive watershed algorithm for the segmentation of soil particles in X-ray three-dimensional microfocus computer tomography images. An erosion-filling-dilation technique is also proposed, which is particularly suitable for processing images of highly porous particles. An s-factor is introduced to improve segmentation quality. In the inverse Euclidean distance map of the binary image, the topography in the zone around the local minima defined by this factor is modified. By using an appropriate value of s, the catchment basins and thus the watershed lines can be more realistically defined and it therefore resolves the problem of ill-segmentation. The proposed methods are applied to assemblages composed of highly porous particles with various size distributions, from uniformly graded to well graded or gap graded. The value of s can be decided based on visual examination of the segmentation quality on a sub-volume of the entire scanned domain. It is found that an s-factor that ranged from 0·5 to 0·7 is applicable to all of the studied cases, which enables a balance between under- and over-segmentation. Particulate-scale information including particle size distribution and particle shape characteristics is evaluated from the processed images and the results are compared with other available methods in the literature. Promising results are obtained.