Title: Modern usage in representation reconstruction methods: an empirical way of GAN to provide solutions for multiple sectors

Authors: Rohit Rastogi; Vineet Rawat; Sidhant Kaushal

Addresses: Department of CSE, ABES Engineering College, Ghaziabad, U.P., India ' Department of CSE, ABES Engineering College, Ghaziabad, U.P., India ' Department of CSE, ABES Engineering College, Ghaziabad, U.P., India

Abstract: Image restoration poses a formidable challenge in the field of computer vision, endeavouring to restore high-quality images from degraded or corrupted versions. This research paper conducts a comprehensive comparison of three prominent image restoration methodologies: GFP GAN, DeOldify, and MIRNet. GFP GAN, featuring a specialised GAN architecture designed for image restoration tasks, introduces an AI-centric approach. DeOldify, a deep learning-based method, focuses on colourising and restoring old images using advanced AI techniques, while MIRNet offers a lightweight network specifically crafted for image restoration within an AI framework. The comparative analysis involves training and testing each method on a diverse dataset comprising both degraded and ground truth images. Employing a confusion matrix, precision, accuracy, recall, and other evaluation metrics are computed to comprehensively assess the performance of these AI-based methods. The matrix affords insights into the strengths and weaknesses of each AI-driven approach, providing a nuanced understanding of their respective performances.

Keywords: GFP GAN; DeOldify; MIRNet; confusion matrix; restoration; image enhancement; computer vision; AI techniques; precision; accuracy; recall; noise; blur.

DOI: 10.1504/IJAMECHS.2025.145726

International Journal of Advanced Mechatronic Systems, 2025 Vol.12 No.2, pp.75 - 92

Received: 11 Mar 2024
Accepted: 03 Sep 2024
Published online: 17 Apr 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article