Research On Shape Reconstruction And Denoising Based On 3D Mesh Data

With the convenience of 3D data acquisition,a large number of 3D data appear,and the demand for improving the quality of raw 3D data is increasingly urgent.Generally,the processing of 3D mesh includes two stages: reconstruction and denoising.That is,3D mesh is reconstructed from 2D observations and then denoised.The 3D mesh reconstruction and denoising algorithm proposed in this paper aims to study how to improve the quality of 3D mesh output in these two stages.Sharp edges are key features that ensure high quality and rich detail of 3D mesh.However,in the past 3D mesh reconstruction or denoising work,it has not attracted enough attention,and is often erroneously treated as noise.For the 3D mesh reconstruction task,we need to solve the problem of reconstruction quality degradation caused by the visual inconsistency between 2D and 3D edge.To this end,this paper,inspired by using edge feature information,designs a graph neural network,driven by edge loss and regularization terms,through the features extracted from a single color image to drive a predefined 3D mesh to produce deformation.Experimental results show that the performance of the algorithm is equivalent to the state-of-the-art algorithm,and the edge visual defect is eased.This shows the effectiveness of the algorithm in improving the quality of 3D mesh in single-view scenario where traditional multi-view 3D reconstruction algorithms are limited.For the 3D mesh denoising task,how to balance the smooth surface and sharp edges of the model is a big challenge.Different from the classical method which focusing on smooth surface,this paper designs a trilateral filter that can perceive features under the framework of energy minimization to address the aforementioned challenge.The problem is modeled as minimizing the energy function of a specific signal on 3D mesh.The data and smooth terms of the energy function are carefully designed.In addition,this paper introduces a trilateral filter that can perceive features.This filter provides a high-quality mesh guidance for the energy model by detecting sharp features on the mesh and distinguishing them from smooth surface.The algorithm in this paper fuses the energy model and filter,incorporates more priori information,and finally obtains better denoising performance after global energy optimization.The denoising experiment on synthetic and real mesh shows that the algorithm's denoising performance surpasses the state-of-the-art algorithms in subjective and objective evaluations.This paper also verifies the robustness and effectiveness of the algorithm through ablation and parameter sensitivity experiments.The robustness and efficiency of the denoising results indicate that our algorithm is suitable for future applications that use high-quality 3D mesh data.