Research On Feature-preserving Restoration Of 3D Data

As a new type of digital media,3D data is widely used in many fields such as restoration of cultural relics,medical diagnosis,digital entertainment,intelligent manufacturing,smart cities,3D printing and so on.With the development of 3D scanning equipment and technology,the acquisition of 3D data has become more convenient and popular.However,in the process of data acquisition,occlusions,mismatches,and jitters are unavoidable,resulting in missing and incomplete data.This makes the reconstructed3 D models have holes,which affects subsequent analysis and editing operations of the 3D models,making the model restricted in various applications.Therefore,3D data restoration is essential.The goal of 3D data restoration is to efficiently and robustly repair defects with 3D data and restore salient features of the data.This research focuses on the 3D data restoration problem with salient features.The main work is as follows:Firstly,a 3D data restoration method based on extended total variation regularization is proposed.First,we identify the hole area of the 3D model and initialize the hole connectivity using dynamic programming method.Then,a variational repair model is proposed based on the complete connectivity and is optimized iteratively by augmented Lagrangian method to get the optimal vertex positions.The experimental results show that the algorithm can effectively recover the features of holes and reconstruct the entire model globally while maintaining the original features of the model.Secondly,a 3D data repair algorithm based on iterative optimization of "connectivity-position" is proposed.Firstly,we identify and initialize the holes.Then,we detect the boundary feature points and fit a feature curve using Bézier curve and adjust the connectivity of the hole area according to the feature curve.Thirdly,a local repair variational framework based on the holes and its neighborhood information is proposed,and solved locally to get the optimal positions of holes and their neighborhood vertices.We iteratively modify the connectivity and optimize vertex positions until the connectivity adjustment no longer occurs.The experimental results show that the algorithm can effectively recover the hole area with missing significant features.Thirdly,a 3D data repair algorithm based on generative-adversarial networks is proposed.The reconstruction error of the 3D model is calculated through the encoding-decoding structure to ensure the consistency of the input and output.We adopt the global discriminator and local discriminator to discriminate the integrity and authenticity of the generated models.The generation network is trained to deceive the two discriminators to generate a locally and globally consistent 3D models.Experimental results show that the method can effectively recover the 3D models with a lot of missing information.Finally,a 3D data repair system is designed and implemented to verify the effectiveness of the proposed algorithms.