Damaged relief and mosaic image inpainting using generative adversarial network

Mosaic and relief works are among the works of art unearthed in archaeological excavations in our country. These artifacts have been influenced by the period in which they were found throughout history, and have been developed and used by many different civilizations. It is possible for materials to wear out and wear out due to their structure. Mosaic and relief works can also come to our day in a worn or damaged form due to natural conditions or the negative effects of human beings. The necessity of repairing the damage in historical artifacts and reaching their original appearance is one of the basic needs in archeology. The problem of image inpainting and creating the original version of the image is a current problem that is tried to be solved with different techniques in the literature. In this thesis, the results of the image inpainting problem by applying the generative adversarial network methods, one of the deep learning-based methods, were examined. Comparative results are given by looking at the structural similarity index, peak signal-to-noise ratio and mean square error metrics of the methods applied in the study. In the study, it was determined that the structural similarity index between the damaged image and the original image on the mosaic and relief data set, with the contextual attention method, provided the best performance with 0.92 - 0.97 in the slightly damaged images and 0.82 - 0.93 in the heavily damaged images. Within the scope of this thesis, answers have been sought to the questions of whether it is possible to repair the damage in historical artifacts with artificial intelligence algorithms and how was the original appearance. In order to reach these answers, firstly, a visual data set of mosaic and relief works in Turkey was created and brought to the literature. This study has the potential to be used as an efficient tool in excavation and museum areas with the rapid visualization of the original image of archaeological artifacts.