© Springer Nature Switzerland AG 2019. Segmentation of tomography images of granular materials has mostly been done without any semantic notion of the materials these images represent. Poor segmentation; undersegmentation or oversegmentation can be mitigated through the introduction of a prior that defines what a valid boundary between two labels should be. In this paper, we present an approach to multiscale tomography segmentation of granular materials, based on hierarchical segmentation using minima. A 3D tomography image of granular materials is segmented with the introduction of a contact prior combined with other extinction for the purpose of formulating the hierarchy. The contact prior based hierarchy is biased towards a segmentation with a realistic mechanically valid contact network. The proposed method is observed to significantly increase the segmentation accuracy of these grains.