9th International Conference on ICT for Sustainable Development

Authors - Tarang Boharupi, S. Poonkuntran
Abstract - Preserving the cultural heritage of iconic landmarks is paramount in maintaining their historical significance. This research focuses on the restoration of damaged images of the Mysore Palace, a heritage site of immense cultural importance. A dataset comprising 171 real and damaged images of the palace, captured with a OnePlus mobile device, forms the basis of this study. The proposed methodology leverages an inpainting technique to generate mask images, which are subsequently utilized in the training of a Generative Adversarial Network (GAN). The GAN architecture is employed to learn the intricate patterns and details necessary for reconstructing the damaged portions of the images. The training process resulted in a model with a PSNR (Peak Signal-to-Noise Ratio) value of 16.5 and SSIM (Structural Similarity Index) of 0.023. The outcomes of this research showcase the efficacy of the developed GAN model in restoring damaged images of heritage structures. The achieved PSNR value indicates a notable improvement in image quality, while the SSIM metric reflects the preservation of structural details. The study contributes to the field of image restoration, especially concerning heritage conservation, offering a novel approach that can be extended to safeguard other cultural landmarks.