Damage Segmentation and Restoration of Ancient Wall Paintings for Preserving Cultural Heritage

Rajasthani paintings are unique and intriguing art forms representing India’s rich cultural heritage. However, these ancient master pieces are deteriorating due to the passage of time, environmental factors, and human actions. Preserving and Restoring these delicate artworks is crucial. One approach to aid their digital restoration is leveraging advanced technologies like deep learning. This study applies image segmentation and restoration techniques to restore the Rajasthani murals in the Mandawa region of rural rajasthan, India. The main objective is to segment the damaged murals, generate their corresponding binary masks and restore the corresponding areas of the damaged image. The research aims to achieve robust and accurate predicted masks for the murals by utilizing state-of-the-art deep learning models and using their outputs as inputs for image restoration as the final restored output image. Extensive comparisons with different segmentation models show that the proposed approach outperforms the rest with an mIOU of 0.892. The proposed method also demonstrates remarkable inpainting results with an SSIM score of 0.9812 on test images. Results show that the method achieves promising restoration of damaged ancient Indian Wall Paintings.