Research And Application On Repairing Method Of Terracotta Warriors Amy Based On Multi-scale Point Clouds And Surface Texture Features

Ancient cultural relics have abundant cultural connotations.Due to natural or man-made factors,the unearthed cultural relics have surface defects,which affects the subsequent work of cultural relics reassembly and virtual display.With the development of 3D(3-Dimension)laser scanning technology,3D digital restoration has become a challenging problem in the field of virtual restoration of cultural relics,which is of great significance for recreating the full appearance of cultural relics.However,in order to improve the efficiency of restoration,the cultural relic model will be downsampled first,and as a result,some local features will be lost.In addition,the surface of the restored three-dimensional cultural relic lacks the information of its surface color and texture.Therefore,this thesis applies deep learning technology to repair the surface geometry and texture feature information from the threedimensional and two-dimensional perspectives.The main research progress is as follows:(1)For the problem of the structural repair of 3D cultural relics,a 3D model completion network based on multi-scale point cloud upsampling GAN(Generative Adversarial Network)network is proposed.Filling the holes with the idea of upsampling the sparse points of the holes as dense points.Using the proposed deep learning network to obtain the repaired multi-scale upsampling point cloud model,and then the hole regions in the multiscale predicted point cloud model are extrated and merged.Finally,the hole patch is merged with the original model to obtain the 3D model completion result.Experimental results show that the up-sampling method can generate a uniform and dense point cloud model,and can be better integrated with the hole neighborhood.(2)Aiming at the problem of the lack of surface texture of cultural relics after 3D repair process,and this problem is transformed into a 2D(2-Dimension)image inpainting problem.A 2D texture repair network based on edge prediction is proposed,and a refined network is added to increase the output image resolution.First,the edge prediction image of the hole area are obtained through the edge generation network of the first stage.Second,combining the edge prediction map and the RGB missing image,and utilizing the coarse network of the second stage to obtain the preliminary RGB repair results.And then,combining the edge prediction image and the coarse inpainting image as the input of the fine network to obtain higher-quality inpainting results.Experiments show that the improved model in this thesis can generate more refined image restoration results.(3)A new framework for digital restoration of 3D cultural relics is designed,which can inpainting the surface color and texture information of cultural relics while repairing the 3D structural features.The framework utilize the network model proposed in(1)to repair the3 D structure information,and then it is packaged as a 3D mesh model.Secondly,the method in(2)is applied to obtain the color and texture information of the cultural relics.Finally,the ZBrush software are employed to map the inpainting 2D image back to the 3D mesh model surface to obtain the final 3D completion result of terracotta warriors.