Image segmentation techniques for granular materials

To improve understanding of the mechanical behavior of granular materials it is important to be able to quantify the relative arrangement of the grains, i.e. the fabric. This can be done, for example, by measuring the orientations of the particles (e.g. the long axis orientation) or by considering the orientations of the vectors normal to each grain-grain contact. In two dimensional (2D) analyses this information can be obtained by digital image analysis of images of thin sections obtained from an optical microscope. While such data is useful, granular materials of engineering interest are three dimensional (3D) materials and quantification of the 3D fabric is necessary. Micro Computed-Tomography (μCT) together with 3D image analysis has emerged as a promising technique for obtaining the 3D data required. This paper aims to highlight the challenges associated with using image analysis to provide quantitative information on fabric. While automated image segmentation has proved to produce reasonable results in some cases, it is sometimes less successful when dealing with highly irregular and angular soil grains. This paper evaluates the effectiveness of 2D and 3D segmentation techniques that rely on the watershed segmentation algorithm. The primary material considered is Reigate Silver Sand, a natural quartzitic sand with grain diameters in the range of 150-300μm. While the sand considered is primarily of interest to geotechnical engineers, the results of this study will be of interest to anyone seeking to quantify granular material fabric using either 2D microscopy data or μCT 3D data sets. © 2009 American Institute of Physics.