Otsu’s Algorithm in the Segmentation of Pore Space in Soils Based on Tomographic Data

Modern analytical research methods tend to automate the process and minimize any manual intervention. This also applies to image analysis of tomographic slices as part of the tomographic study of soils. To calculate morphometric parameters, a tomographic image must be segmented (divided into phases) via automated or manual thresholding. The problem in automated thresholding algorithms is insufficient accuracy when dealing with different data. The goal of this study is to apply one of the most common Otsu’s thresholding algorithms to tomographic images of different soils, to show how justified is its use, and to determine the causes and conditions of errors in automated segmentation. The research has been performed at the Laboratory of Soil Physics and Hydrogeology with the Dokuchaev Soil Science Institute. We have used the tomographic images of urban soils (Urbic Technosols), dark-gray soil (Chernic Phaeozem), and soddy-podzolic soil (Albic Retisol) scanned with different equipment. The automated segmentation results are compared to manual thresholding. The pore space of soils is used as the control X-ray contrast phase and the porosity and the amount of pores, as benchmarks. The results show that Otsu’s method most accurately works with large data, when the image artifacts (digital noise) are minimal or absent at all, which is a typical situation for the aggregates <1 mm. As for coarse-resolution surveys or noisy images typical of samples with a high X-ray absorption, automatic segmentation is highly undesirable.