Quantifying carbonate microstructure using classical segmentation pipelines for CCS and radioactive waste applications

Abstract. Digital-rock analysis can now simulate pore-scale flow and geomechanics with high fidelity, yet routine application is still limited by the cost of converting images into pixel-accurate pore masks. Some porosity measurements measure total porosity, but flow requires connected porosity. The need to distinguish between isolated and connected pores is clearly important. The question is why it is difficult to identify connected and unconnected pores. To tackle this, we present a fully labelled 2-D scanning-electron-microscope (SEM) image dataset of outcrop carbonates designed to separate isolated intragranular pores from the connected fracture–pore network. This distinction controls seal performance in carbon-capture and storage (CCS) and radioactive-waste repositories. Each of the 29 056 × 22 952 px images were partitioned into 100 tiles (2048 × 2048 px), annotated by polygon tracing. We then employ and benchmark eight unsupervised computer-vision algorithms: Morphological Gradient, Distance Transform, Local Contrast, Watershed, Global Threshold, Gabor Texture, Refined Morphology, and Edge-based methods. We contribute (i) an open carbonate SEM dataset with labels separating connected and isolated pores, (ii) an efficient polygon-based labelling workflow, and (iii) a quantitative comparison of eight classical pipelines plus a Hybrid voting ensemble. These resources shorten the path from raw images to pore-network predictions, and provide a foundation for future learning-based methods.