Research On Murals Inpainting Based On Generative Adversarial Networks

Today,with the rapid development of human society,the rise of culture has become an inevitable choice for the growing comprehensive strength of a country.The murals of Mogao Grottoes in Dunhuang have a large scale,rich and magnificent contents and a long history.They are the unity of aesthetic,historical and artistic values,and the world cultural heritage of civilization at home and abroad.However,due to the accumulation of natural disasters and other irresistible factors and human factors,murals have been seriously damaged.Many diseases such as peeling off of the wall still threaten the safety of murals.Since the introduction of relevant national policies on cultural relics protection has basically eliminated man-made destruction,but the destruction of natural factors still exists,which is an urgent problem to be solved in the work of cultural relics protection.Through the digital image restoration technology,the damaged Dunhuang murals can be protected and restored non-contact,which is of great significance for the restoration of Dunhuang murals.The restoration speed is faster than the manual restoration,and is the research focus in the field of cultural relics restoration.In this paper,the protection of Dunhuang murals is mainly aimed at the restoration of the defect area,in order to solve the problem of Dunhuang mural damage in the natural state.This paper designs and improves an image repair method based on the generation antagonism network to improve the repair efficiency of Dunhuang mural repair work and improve the accuracy of the repair results.First of all,through the comparative experiment of the current popular image restoration methods,it is determined to use the anisotropic diffusion denoising technology,through the reappearance of the anisotropic diffusion method,the denoising work before the restoration of Dunhuang murals in the natural state is carried out;secondly,this paper designs and improves the network structure of a digital image restoration model based on the generated antagonism network,and through the actual implementation Finally,the repair effect of the damaged mural is improved by improving the loss function of the network and adding the loss function of perception.This paper takes the murals of Mogao Grottoes in Dunhuang as the main restoration object,and realizes the effective restoration of both the murals with large area disaster and small area disaster with small cracks.PSNR and SSIM are used to evaluate the restoration results.The experimental results show that the method in this paper can repair the damaged murals to generate realistic missing area images.