The increase in the application of unconventional resources and the presence of heterogeneity in such reservoirs have increased the need to consider multi-scale models. Some of the recent studies indicate that sub-resolution porosity (SRP) has a profound influence on the flow characteristics of porous. Thus, the objective of this work is to present a technique to restore lost details from images of porous media. Interpolation and filtering are some of the conventional techniques that have long been used to improve image resolution. However, this field has changed with the current development of artificial intelligence, especially generative adversarial networks (GAN). GANs consist of two neural networks: A generator and a discriminator, which are trained in an adversarial manner to produce realistic images. When applied to petroleum engineering, GANs are employed in super-resolution tasks, in which the GANs learn to reconstruct high-resolution images from low-resolution inputs. In particular, a synthetic dataset of low-resolution images is generated from an existing dataset of a sandstone sample. In this research, we employ RealESRGAN which is more effective than previous models such as SRGAN. Finally, we constructed a pore network based on generated images and the model's results were almost identical to the actual model.