Computer Aided Mural Inpainting Based On Generative Adversarial Networks

Murals are a precious human cultural heritage.However,due to various natural and manmade reasons,most of the current murals have suffered varying degrees of damage.In order to inherit these precious historical and cultural heritages,it is necessary to repair and protect the existing murals.With the development of computer technology,the use of computer technology to assist in the inpainting of murals can not only save manpower and material resources,but also avoid human factors which may cause secondary damage to murals.Aiming at the problem that the existing computer-aided mural repair algorithm cannot effectively repair the large-area defects of murals,this paper uses the generated confrontation network to conduct the experimental study on the repair of large-area defects in murals.The main work includes the following four points: 1.For the first time,the generative adversarial network is introduced into the field of mural inpainting.Through the powerful learning ability of the artificial neural network,the overall structure of the mural is used to effectively repair the missing area structure.And the algorithm is verified to repair the large damaged area better,compared to the traditional way,which achieves the purpose of repairing with the overall information of murals.2.Aiming at the shortcomings of the existing network structure in the repair of mural defects,an improved generative adversarial network structure is proposed,and the effectiveness of the improvement of each module of the network is verified.Compared to the existing mural repair algorithm,the peak signal-to-noise ratio(PSNR)score increases by 4%,and the structural similarity(SSIM)score increases 2%.3.Aiming at the problem that the color of faded murals is difficult to recover,the mural defect repair network structure proposed in this paper is adjusted to further restore the color of mural images,which provides an effective way for color restoration of murals with color degradation.4.According to the problem of the existing mural datasets,this paper created a dataset of Dunhuang mural repairs containing 855 images through the collection,collation and defect simulation of Dunhuang murals.In summary,the mural inpainting algorithm proposed in this paper can repair the defect area and color of the mural more effectively and provide reference for the restoration work of mural painting,which promotes the development of mural protection.