Abstract:
X-ray computed tomography (XCT) is one of the key methods for studying the internal structure of porous samples, such as oil and gas bearing reservoir rocks. To model the properties of porous media based on XCT data, these images must be segmented. However, the correctness of the segmentation procedure cannot be verified due to lack of ground-truth data. This paper describes the development, testing and approbation of software for creating synthetic tomograms of porous media to solve the problem of the lack of validation or training segmentations. The created framework is based on the direct and inverse Radon transform. To test the proposed approach, we compared the speed and quality of key functions with existing analogues. Based on proxy samples obtained by segmenting XCT images, we created synthetic images of porous media from five phases: pore (air with negligible attenuation), kaolinite (Al$_2$Si$_2$O$_5$OH$_4$), silicon dioxide (SiO$_2$), calcium carbonate (CaCO$_3$), and iron disulfide (FeS$_2$). There is good agreement between the obtained synthetic data and the original XCT images. The developed technique allows solving the problem of creating labeled data for using machine learning in CT image segmentation tasks, as well as for testing any other methods for CT image segmentation.