Accurately integrating pore geometry analysis into velocity predictions remains a challenge in heterogeneous formations, such as Brazilian Pre-Salt carbonates, where complex pore structures significantly impact petrophysical properties. This study aims to evaluate the petrophysical properties of carbonates from the Santos basin, offshore Brazil. We employed an advanced methodology using machine learning to reconstruct the S-wave velocity ( \(V_S\) ) log from P-wave velocity ( \(V_P\) ), density ( \(\rho _b\) ), and neutron porosity ( \(\phi _N\) ) logs. Among the models tested, the Multilayer Perceptron (MLP) architecture demonstrated the highest accuracy, achieving the lowest error values. Additionally, visual computing was employed to extract geometric characteristics related to porosity from mCT images to apply Kuster-Toksöz (K-T) and Self-Consistent (S-C) theoretical effective models to estimate elastic velocities. Our results showed that the K-T model tends to overestimate velocity values, while the S-C model yields values closer to the experimental ones. A significant relationship was observed between pore aspect ratios and the velocity values, emphasizing the impact of the pore geometry on the velocity predictions. Furthermore, it was noted that analyzing two-dimensional images can introduce variations in these measurements. This integrated approach not only enhances the understanding of reservoir behavior, but also introduces a novel application of computer vision techniques in geophysical studies.
