TY - JOUR

T1 - Variational-based segmentation of bio-pores in tomographic images

AU - Bauer, Benjamin

AU - Cai, Xiaohao

AU - Peth, Stephan

AU - Schladitz, Katja

AU - Steidl, Gabriele

N1 - Funding information: GS and KS were partly supported by the German Federal Ministry of Education and Research through project 05M13 (AniS). The authors thank Prakash Easwaran for the simulated bio-pore systems.

PY - 2017/1

Y1 - 2017/1

N2 - X-ray computed tomography (CT) combined with a quantitative analysis of the resulting volume images is a fruitful technique in soil science. However the variations in X-ray attenuation due to different soil components keep the segmentation of single components within these highly heterogeneous samples a challenging problem. Particularly demanding are bio-pores due to their elongated shape and the low gray value difference to the surrounding soil structure. Recently variational models in connection with algorithms from convex optimization were successfully applied for image segmentation. In this paper we apply these methods for the first time for the segmentation of bio-pores in CT images of soil samples. We introduce a novel convex model which enforces smooth boundaries of bio-pores and takes the varying attenuation values in the depth into account. Segmentation results are reported for different real-world 3D data sets as well as for simulated data. These results are compared with two gray value thresholding methods, namely indicator kriging and a global thresholding procedure, and with a morphological approach. Pros and cons of the methods are assessed by considering geometric features of the segmented bio-pore systems. The variational approach features well-connected smooth pores while not detecting smaller or shallower pores. This is an advantage in cases where the main bio-pores network is of interest and where infillings, e.g., excrements of earthworms, would result in losing pore connections as observed for the other thresholding methods.

AB - X-ray computed tomography (CT) combined with a quantitative analysis of the resulting volume images is a fruitful technique in soil science. However the variations in X-ray attenuation due to different soil components keep the segmentation of single components within these highly heterogeneous samples a challenging problem. Particularly demanding are bio-pores due to their elongated shape and the low gray value difference to the surrounding soil structure. Recently variational models in connection with algorithms from convex optimization were successfully applied for image segmentation. In this paper we apply these methods for the first time for the segmentation of bio-pores in CT images of soil samples. We introduce a novel convex model which enforces smooth boundaries of bio-pores and takes the varying attenuation values in the depth into account. Segmentation results are reported for different real-world 3D data sets as well as for simulated data. These results are compared with two gray value thresholding methods, namely indicator kriging and a global thresholding procedure, and with a morphological approach. Pros and cons of the methods are assessed by considering geometric features of the segmented bio-pore systems. The variational approach features well-connected smooth pores while not detecting smaller or shallower pores. This is an advantage in cases where the main bio-pores network is of interest and where infillings, e.g., excrements of earthworms, would result in losing pore connections as observed for the other thresholding methods.

KW - 3D image segmentation

KW - Bio-pores

KW - Root system

KW - Variational segmentation

KW - Total variation minimization

KW - Gray value thresholding

KW - Morphological segmentation

UR - http://www.scopus.com/inward/record.url?scp=84991704650&partnerID=8YFLogxK

U2 - 10.1016/j.cageo.2016.09.013

DO - 10.1016/j.cageo.2016.09.013

M3 - Article

VL - 98

SP - 1

EP - 8

JO - Computers & geosciences

JF - Computers & geosciences

SN - 0098-3004

ER -