Image processing of high-resolution 3D images to create digital representation of pore microstructures for image-based rock physics simulations remains a highly subjective enterprise, despite the seemly precision associated with improving imaging resolutions and intensive parallel computations. The decisions on how to identify pore space, both macro- and micropores, and various mineral components remain very much dependent upon user choices and biases. This study demonstrates how uncertainty can be quantified for a highly subjective segmentation process. A set of shaly sand samples with significant amounts of authigenic chlorite/smectite that lines larger pores was tested to identify uncertainty quantification (UQ) requirements associated with image-processing steps, segmentation in particular. Much of the porosity in these coarse-grain samples is associated with subresolution micropores that complicates their assignment in any pore-grain segmentation strategy. Two segmentation strategies, a binary segmentation with a linear-threshold and a machine-learning (ML) approach to two-phase segmentation, are employed with different UQ parameter space. The contribution of resolvable macropores in these samples, and their spatial distributions with regard to pore-lining clay mineral with unresolvable microporosity, are iteratively studied over the defined UQ parameter space, and cross-validated by independent NMR and MICP measurements. The pore structure extracted from these different iterations was the basis of simulations for basic petrophysical properties. Upon cross-validation of simulated results with measured core properties, a UQ framework is proposed to assess the differences between the different measurements from three angles: sampling, numerical and physical.