Separation of sand and gravel particles in 3D images using the adaptive h-extrema transform

Size and shape properties of particles captured by micro-computed tomography are useful for optimization of packing density or virtual modelling of granular systems. Scanning samples as packed beds without sample preparation allows rapid capturing. However, separation by image processing is then required for particles in contact. The commonly applied separation strategy based on watershed transform on the inverted distance map requires users to choose a smoothing algorithm and its parameters, usually a time-consuming trial and error procedure. Here, the adaptive h-extrema transform is chosen, useful for separating multi-sized particle samples. Features available before separation are used to estimate its input parameters and analyse the correlation to the resulting level of correctly separated particles. As a result, an automatic procedure is proposed with input parameters estimated from packing density. Additionally, the influence of contact on measurement accuracy of size properties is analysed indicating that physical separation is not necessarily required.