Improving Core CT Scan Image Quality Using Super Resolution Techniques an‎d Extracting Core Morphological Properties

10/6/2026 12:00:00 AM

High-resolution micro-CT imaging of porous reservoir rocks plays a crucial role in digital rock physics an‎d fluid flow simulation in porous environments. However, the simultaneous limitation of achieving high resolution an‎d a wide field of view remains one of the primary challenges in this field. In this study, a Generative Adversarial Network (GAN)-based image resolution enhancement framework is proposed, trained on high-resolution an‎d denoised CT slices, with the aim of accurately reconstructing fine-scale pore structures from low-resolution initial images. The developed model is compared with seven other convolutional neural network (CNN)-based methods as well as the Bicubic interpolation (BICUBIC) method. This comparison is carried out using a comprehensive set of eva‎luation metrics, including image-based indices (such as PSNR, SSIM, LPIPS) an‎d petrophysical indices (such as porosity, connectivity, pore size distribution, skeleton to pore ratio, an‎d grain boundary preservation). The results indicate that while CNN-based models exhibit high pixel accuracy, the GAN model provides superior perceptual quality an‎d reconstructs physical structures that are much closer to the ground truth (GT) data. Moreover, numerical flow simulations at the pore scale on 3D core images demonstrate that the GAN-generated outputs can accurately reproduce reservoir properties such as relative permeability, diffusion coefficient, an‎d relative diffusivity, significantly outperforming classical interpolation method. The findings of this research show that the proposed GAN model not only improves the perceptual quality of images but also preserves critical petrophysical features, making it an effective tool for simulation analyses an‎d reservoir studies.

سوپررزولوشن , شبكه‌هاي مولد تخاصمي , تصاوير ميكروسي‌تي مغزه , خواص پتروفيزيكي , شبيه‌سازي جريان حفره‌اي

Super Resolution , Generative Adversarial Networks (GANs) , Micro-CT Core Images , Petrophysical Properties , Pore-Scale Flow Simulation

Mohammad Hamidian

Rohaldin Miri / Hossein Fazeli