Paper Details

Name: Santosh Dattatray Kale, Shailendrakumar M.Mukane
Country: India

Year: 2026
Volume: Volume No: 13, January, Year: 2026 (Special Issue)
Page Number: 958-968
DOI: https://doi.org/10.5281/zenodo.20624339

Image inpainting has undergone a transformative evolution in recent years, propelled by the rise of deep learning. This review provides a critical analysis of the latest advances in image inpainting, focusing on methods grounded in Generative Adversarial Networks (GANs), Convolutional Neural Networks (CNNs), Transformers,
and Diffusion Models. We organize the literature across algorithmic and architectural lines, highlight functional innovations such as multi-scale attention, structure-texture separation, and pseudo-labelling, and evaluate their performance using metrics like PSNR, SSIM, LPIPS, and FID on standard datasets. In addition, we explore domain-specific challenges, particularly high-angle fractures and periodic defects, and offer insight into future research opportunities. This study serves as a comprehensive guide to the current landscape and evolving frontiers in deep learning-based image inpainting.