Synthesis of Sphere Packings for Evaluation of Image Segmentation Algorithms.

X-ray CT provides powerful means for characterization and visualization of phase distributions in porous media. However, selection of appropriate segmentation algorithms for conversion of fuzzy grayscale images to distinct phases is challenging, mainly because of the lack of ground truth data. To overcome these problems, we first applied Discrete Element Modeling to generate realistic sphere packings exhibiting different porosities and particle size distributions. Then lattice Boltzmann simulations were conducted to generate wetting phase distributions for various saturation levels. Finally, different noise sources and levels that resemble noise introduced during X-ray CT scanning were added to test the applicability and potential failure of global and locally adaptive two- and multi-phase segmentation algorithms.