Ancient Indian Murals Digital Restoration through Image InPainting

Murals and paintings are an integral part of the history of any civilization. India has deep roots in ancient history and possesses some of its most valuable assets in ancient murals dating back several hundred years. They feature Hindu mythology, traditions, incidents of bravery and cultural events and thus have a substantial social significance in human society. Unfortunately, some of these assets lose value over time due to environmental factors and human negligence. The ground truth regarding their original colour, textures and painting style references is not currently available, due to which their physical restoration is not possible. Therefore, the only way to restore them is through their digital counterparts and image inpainting techniques. In this paper, a complete end-to-end image restoration framework has been presented to virtually repair the damaged regions of the mural paintings using a selective neural style technique for the transfer of realistic-looking damaged textures, a modified U-Net for detecting the damaged regions in an image and a Partial Convolution network for effective image inpainting. The photographs of Rajasthani paintings scrapped from online sources are used as the dataset. The restored results based on the perceptual quality of completion on ancient Indian murals are promising; thus, they may contribute to the digital restoration of murals.