Research On Denoising And Registration Of Surface Model

The advantages of the three-dimensional model are that it can not only visually reproduce the objects in the real world,but also fully describe the geometric characteristics of the objects,which is impossible to achieve with two-dimensional images.Therefore,in recent years,the analysis and research of 3D models have been endless.However,no matter how high the accuracy of the data collection equipment,a lot of noise will inevitably be generated during the digitization process.These noises not only reduce the quality and visual effect of the model,but also produce analysis and processing of the modeled 3D model.Great influence;In addition,a large number of images cause inconsistency in image information due to differences in imaging equipment,environment,timing,location,etc.How to eliminate the differences and obtain consistent information has become an urgent problem to be solved.Therefore,it is of scientific significance to study the denoising and registration algorithms of 3D models.The main research work of the thesis is as follows:(1)In terms of denoising,this thesis designs a scheme to retain geometric features(especially sharp features and shallow features)while removing noise.Given the input of the noise model,first use the improved isotropy The method filters the vertices,and then adds the anisotropic L0 filter to strengthen the feature.Then,a modified joint bilateral filtering method is used to process the normal vector field of the input model,and finally the vertex position is updated with the filtered surface normal.A large number of experiments on various noise models have proved the effectiveness of this method in retaining sharp(high curvature)and shallow(low curvature)features.Compared with other denoising methods,this thesis can save more smooth shallow.The geometric details.In addition,the qualitative and quantitative comparisons further show that the proposed method has good performance in recovering the geometric shape of a given model,and also verifies the robustness of the algorithm.(2)In terms of registration,a coarse and fine cascade registration algorithmbased on tensor voting is proposed.First calculate the voting tensor of the vertex of the 3D model to be registered,classify it into feature points and non-feature points,and then use the fast 4PCS algorithm to implement the coarse registration of the model on the feature point set,reducing the algorithm screening and verification time,Improve the calculation efficiency;finally,by adding random screening method,nearest neighbor method,uniformly assigned weights and fixed ratio method iteration to the ICP algorithm to improve the noise resistance and convergence speed of the algorithm,thereby achieving the final accurate matching of the two surfaces quasi.Experiments show that the algorithm in this thesis is a high-precision and strong anti-noise 3D model registration method.