Segmentation of X-ray microtomography images of soil using gradient masks

For many analyses, grey scale images from X-ray tomography and other sources need to be segmented into objects and background which often is a difficult task and afflicted by an arbitrary and subjective choice of threshold values. This is especially true if the volume fraction of objects is small and the histogram becomes unimodal. Bi-level segmentation based on region growing is a promising approach to cope with the fuzzy transition zone between object and background due to the partial volume effect, but until now there is no method to properly determine the required thresholds in case of unimodality. We propose an automatic and robust technique for threshold selection based on edge detection. The method uses gradient masks which are defined as regions of interest for the determination of threshold values. Its robustness is analysed by a systematic performance test and finally demonstrated for the segmentation of pores in different soils using images from X-ray tomography.