A novel local-minima 3D image segmentation method for fluid flow in low-resolution porous material images

Image segmentation is a prominent process in Digital Rock Physics analysis. It directly affects the accuracy of the fluid-flow simulation in porous mediums. The accuracy of conventional global thresholding segmentation depends on image resolution. Coarse-resolution digital rock images are normally used to avoid high computation costs in processing high-resolution images that usually have
more accurate results. But the image segmentation implemented on the low-resolution images becomes more arbitrary and thus generates less satisfying simulation results. This work develops a novel local-minima 3D image segmentation method that can improve the accuracy of simulations of fluid flow in low-resolution rock images. It adopts two global threshold values to capture the convinced pure pore and solid phase. Voxels having greyscale values between the two thresholds are assigned to a temporarily uncertain phase. A search algorithm is then applied to find the local minima in the uncertain region. These local minima are pores while the rest are solids. Indiana Limestone and Bentheimer Sandstone digital rock images scanned at different resolutions are studied to validate the local-minima segmentation method. We apply the conventional global thresholding and the proposed method to these digital models and study the pore- and throat-size distributions by extracting pore networks by a maximal-ball algorithm. The result shows that the local-minima method yields networks that are more accurate than those generated by the global thresholding method. In addition, we calculate the permeabilities of these models by a Lattice-Boltzmann method. The localminima segmentation method yields an average absolute permeability error of 23% for the Indiana Limestone and 13% for the Bentheimer Sandstone, whereas the global thresholding method yields 202% and 83.67% errors, respectively. The result demonstrates that the carbonate rock is impacted more by the coarsening of image resolutions. Our segmentation technique can improve the overall accuracy of fluid flow simulations in low-resolution digital images.