Research And Application Of Digital Generation Technology Of Irregular Fragments Of Tomb Murals

The "Polo" mural of the Tang Dynasty was painted in a narrow tomb passage,making extracting the fragments in sections necessary.However,this process could cause damage and affect the beauty and integrity of the mural.To protect cultural heritage and facilitate restoration,experts have used "re-pasting restoration." However,the small and fragile fragments present limited texture information,making re-pasting difficult and risky.So,this paper proposes a multi-level irregular extension texture generation network that generates texture information around the mural fragments using a non-contact approach,providing restoration experts with a reference and reducing re-pasting risk.The network generates murals with rich and precise textures,achieving good results in texture structure and artistic style.Additionally,a mural irregular extension texture generation system is designed to provide restoration experts with a convenient display,evaluation,and analysis platform,integrating theory and practical application.The main work content is as follows:(1)This paper proposes a multi-level irregular extension texture generation network model that generates textures from the inside out using internal information of irregular fragments in the tomb mural.The model consists of two generative networks with a progressive relationship.The structural generation network generates the external extension information structure of the mural,while the texture generation network refines the texture details,gradually improving the realism of the generated mural.Introducing the extension texture shift layer improves the accuracy of texture generation in irregular regions,obtaining more high-frequency textures.The network also includes skip connections to accelerate the training process and adds dilation convolution to increase the receptive field.The multi-scale discriminator comprises a global and extension texture discriminator network to improve the semantic and texture similarity between the original mural and the generated extension.Optimized content reconstruction loss,texture style loss,and adversarial loss are used to improve the accuracy of mural texture generation while enhancing texture style consistency and content uniformity.Experimental analysis confirms that the proposed multi-level irregular extension texture generation model generates more realistic textures with smoother boundaries for mural fragments.(2)This paper designs the mural irregular extension texture generation system.Using the proposed multi-level irregular extension texture generation network model to generate irregular extension textures for loaded mural fragments.The system also includes functions for displaying evaluation metrics,expert commentaries,and exporting generated murals.The system provides a platform for restoration experts to analyze and communicate mural restoration progress,improving their understanding of mural texture and composition.The successful generation of irregular extension textures for loaded murals using the proposed model demonstrates the feasibility and effectiveness of the model.