Generative Adversarial Networks for Image Restoration and Inpainting
DOI:
https://doi.org/10.15330/itee.2025.3.05Keywords:
generative adversarial network, image inpainting, graphic content, pythonAbstract
The aim of this work is to investigate methods of Generative Adversarial Networks (GANs) for image inpainting tasks and to develop a software implementation of a system for restoring graphic content using enhanced generation algorithms. The research is based on experimental modeling of a neural network and the development of a specialized loss function. The study proposes an improved image generation method with a unique loss function, which reduces generator loss compared to the basic GAN architecture and enhances the quality of graphic content restoration. A prototype of the image inpainting system was developed, demonstrating effective object removal and restoration of image textures and color features. The system provides high processing speed and stability of results, which has been confirmed experimentally. The developed system has practical significance for all areas where visual data processing is critical, including computer vision and information technology. The implementation of the proposed approach improves the quality of object removal in images and expands the capabilities of generative models in tasks of restoration and editing of graphic content.
References
Z. Chen, “Generative Adversarial Networks for Image Restoration: Revolutions, Challenges and Future Look,” Theor. Natural Sci., vol. 151, no. 1, pp. 202–210, Dec. 2025. doi: https://doi.org/10.54254/2753-8818/2026.ch30897.
C. Dong, H. Liu, X. Wang, and X. Bi, “Image inpainting method based on AU-GAN,” Multimedia Syst., vol. 30, no. 2, Mar. 2024. doi: https://doi.org/10.1007/s00530-024-01290-3.
Z. Liu and M. Qin, “Research on image inpainting methods based on machine learning,” Appl. Comput. Eng., vol. 19, no. 1, pp. 67–74, Oct. 2023. doi: https://doi.org/10.54254/2755-2721/19/20231009.
A. B. Yildirim at al. “Inst-Inpaint: Instructing to Remove Objects with Diffusion Models.” arXiv.org. doi: https://doi.org/10.48550/arXiv.2304.03246.
H. Zheng, Z. Lin et al. “CM-GAN: Image Inpainting with Cascaded Modulation GAN and Object-Aware Training.” arXiv.org. doi: https://doi.org/10.48550/arXiv.2203.11947.
R. Wei et al. “OmniEraser: Remove Objects and Their Effects in Images with Paired Video-Frame Data.” arXiv.org e-Print archive. URL: https://arxiv.org/html/2501.07397v3.
A.-A. Barglazan, R. Brad, and C. Constantinescu, “Image Inpainting Forgery Detection: A Review,” J. Imag., vol. 10, no. 2, p. 42, Feb. 2024. doi: https://doi.org/10.3390/jimaging10020042.
“What is a GAN? - Generative Adversarial Networks Explained - AWS.” Amazon Web Services, Inc. URL: https://aws.amazon.com/what-is/gan/
Liu, X., Hay-Man Ng, A., Lei, F., Zhang, Y., and Li, Z., “GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials”, arXiv e-prints, Art. no. arXiv:2209.15454, 2022. doi: https://doi.org/10.48550/arXiv.2209.15454.
Ian Goodfellow, Yoshua Bengio, Aaron Courville. Deep Learning. MIT Press, 2016. URL: http://www.deeplearningbook.org.
Yuqing Ma et al. “Coarse-to-Fine Image Inpainting via Region-wise Convolutions and Non-Local Correlation.” Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence Main track. Pages 3123-3129. doi: https://doi.org/10.24963/ijcai.2019/433.
Линовський, А. О., Мухін, В. Є. “Засоби покращення якості та знешумлення зображень на основі застосування згорткових та рекурентних нейронних мереж.” Телекомунікаційні та інформаційні технології. 2023. № 1(78). С. 82–89. doi: https://doi.org/10.31673/2412-4338.2023.018289.
D. Berdnyk and D. Peleshko, “Image reconstruction using generative neural networks”, Herald of Khmelnytskyi National University. Technical sciences, vol. 325, no. 5(1), pp. 30–34, Oct. 2023, URL: https://heraldts.khmnu.edu.ua/index.php/heraldts/article/view/452.
A. Іванов, В. Онищенко, “Методи генерації зображень з використанням мереж GAN,” Адапт. системи автомат. упр., vol. 1, no. 42, pp. 153–159, May 2023. doi: https://doi.org/10.20535/1560-8956.42.2023.279109.
М. В. Семаньків and О. В. Ціхун, “Вдосконалення методів імпейнтингу на основі генеративних моделей,” Вісн. Східноукр. нац. ун-ту ім. Володимира Даля, no. 8(294), pp. 5–10, Oct. 2025. doi: https://doi.org/10.33216/1998-7927-2025-294-8-5-10.
Downloads
Published
How to Cite
Issue
Section
