Research On The Similarity Analysis And Denoising Of 3D Models Based On Tensor

As a 3D object representation,3D mesh model is widely used in computer graphics,computer aided design and other fields.With the development of 3D technology,3D mesh model can be obtained by 3D scanner or some other modeling tools,so that we can further process 3D mesh models.Among them,3D model similarity analysis and denoising is a hot research topic in digital geometry processing.The research work and the results are as follows:1)Building a sub-model block for similarity analysis and denoising.The similarity analysis method and denoising method proposed in this paper are based on the sub-model block processing.Therefore,for the similarity analysis method,we use the over segmentation algorithm and the clustering algorithm to divide the 3D model into several segmentation blocks.For the three-dimensional denoising method,considering the denoising of the segmentation block boundary points,we sort the obtained points in the sub-model block composed of these points and their neighborhoods.2)A 3D model similarity analysis method based on tensor is proposed.In order to analyze the similarity between sub-model blocks from different features,we use Gauss descriptors,shape index and shape diameter function to construct three order tensors,and compare the similarity of model blocks by constructing the tensor distance.The experimental results show that this method of 3D model similarity analysis can well measure the similarity between sub-model blocks and also have better robustness for different poses of the same model.3)A 3D mesh model denoising method combining tensor and sub-model blocks is constructed.It firstly sets the neighborhood of each vertex as a unit,and finds blocks which have the most similar geometric features.Then it constructs the normal vector of each reference block and similar blocks as a tensor,and finds out the revised normal direction through the tensor eigen decomposition.Finally,it moves the corresponding vertex along the normal direction for a given distance to obtain the denoised position.The experimental results show that the 3D denoising method in this paper can keep the geometric features of the model while reducing the noise.Finally,this paper summarizes the research content and looks forward to the future work.