Investigating the effect of the color spectrum of input images in the process of increasing image quality using a new method based on adversarial generative networks

ElahiKhaledi, Alireza; Jamshidi, Saeid; Masihi, Mohsen

doi:10.22078/pr.2025.5735.3550

Investigating the effect of the color spectrum of input images in the process of increasing image quality using a new method based on adversarial generative networks

Document Type : Research Paper

Authors

Alireza ElahiKhaledi ¹

Saeid Jamshidi ²

Mohsen Masihi ²

¹ Department of Chemical and Petroleum Engineering, Sharif University of Technology, Tehran, Iran

² Department of Chemical and Petroleum Engineering, Sharif University of Technology, Tehran, Iran.

10.22078/pr.2025.5735.3550

Abstract

Reconstructing high-resolution images from poor-quality images is one of the basic challenges in image processing. In modeling porous media using images obtained from computed microtomography (Micro-CT), high-quality images of these media are usually unavailable for different reasons, including the cost and computational complexity of high-quality imaging. With the advancement in the development of multiscale networks, it is possible to use more image details in these networks, and there is a further need for high-quality images. Today, adversarial generative networks are used as a practical tool in increasing image quality. These networks are trained using pairs of high-quality and low-quality images and then, they produce high-quality images by taking low-quality images as input. The final images used in modeling the porous media are binary images, but using binary images as input in the image quality enhancement process may result in detail loss. For this purpose, this study investigates the effect of using grayscale or binary images as input in the model training process. Grayscale images of a Brea sandstone sample are taken as the main input and binary images are generated using grayscale images and the realization of the real porosity of the rock sample. In this research, a new model called RealESRGAN is used to enhance the quality of the images.

Keywords

Generative Adversarial Network (GAN)

Real ESRGAN Model

Grayscale Image

Binary Image

Pore Network Modeling

Subjects