Research On 3D Mesh Model Segmentation Method Based On Wave Kernel Feature And Gaussian Mixture Model

In the research and analysis of 3D mesh model,3D mesh model segmentation is a more common method.With the development of 3D acquisition technology,3D model segmentation algorithm has been paid more and more attention in virtual reality technology,animation technology,biology,medical image scanning and other application fields?Currently,clustering has gradually become the mainstream method of 3D mesh model segmentation,and machine learning and deep learning are gradually integrated into 3D mesh model segmentation.In this trend,the following problems still exist in 3D mesh model segmentation.Firstly,the feature descriptors used to describe the surface features of the model are easily affected by the changes of the model attitude,which makes the calculation error and is not conducive to the accurate segmentation results;Secondly,in the 3D mesh model segmentation method based on machine learning and deep learning,it is difficult to adjust the parameters,which affects the segmentation effect,and the long training time of the model leads to the long time of the whole segmentation process.In order to solve the above problems,based on the existing theories,this paper proposes two new 3D mesh model segmentation methods,which are manifold harmonic basis and wave kernel feature;The clustering generated by Gaussian mixture model is used to segment the 3D mesh model.The efficiency and effect of the two methods are improved.The main work of this paper is as follows:(1)The 3D mesh model is segmented by manifold harmonic basis and wave kernel feature.Wave kernel feature is a feature descriptor in 3D mesh model.In this paper,the manifold harmonic basis of 3D mesh model is calculated by Laplace operator of discretizing 3D mesh,and then the manifold harmonic basis is decomposed to obtain the required eigenvalues and eigenvectors.By using the improved wave kernel feature formula to calculate the wave kernel eigenvalues of each3 D mesh model vertex,the wave kernel eigenvalues of each vertex are substituted into the persistent clustering algorithm as parameters,and the final clustering label is obtained,and the final segmentation result is generated by the clustering label.From the experimental results and the evaluation results of the algorithm,compared with other segmentation algorithms,this algorithm improves the efficiency and accuracy of 3D mesh model segmentation.(2)The improved Gaussian mixture model and EM algorithm are used to segment the 3D mesh model.Gaussian mixture model and EM algorithm are applied to 3D model segmentation.Because the initial parameters of Gaussian mixture model are unstable,this paper first uses Kmeans clustering to get the initial clustering parameters of Gaussian mixture model,that is,the initial mean value of Gaussian mixture model.In this way,the Gaussian mixture model is improved,and then the mean value is calculated iteratively according to EM algorithm,Variance and weight as well as the final clustering label.Finally,the 3D model is segmented according to the clustering label,and the segmentation result of 3D mesh model is obtained.In the experimental process,it is found that the number of clusters has a great influence on the segmentation results,so two methods are proposed to determine the optimal number of clusters,which are elbow method and maximum contour coefficient method.The segmentation number determined by these two methods can achieve more accurate segmentation of 3D mesh model.The contributions and innovations of the two 3D mesh model segmentation methods studied in this paper are as follows:(1)this paper extends the application of flow shaping and harmonic basis,combines the improved wave kernel features and persistent clustering,and solves the problems of low segmentation efficiency and inaccurate segmentation in some existing 3D mesh model segmentation methods.(2)In this paper,we use the clustering center obtained by K-means algorithm as the initial mean parameter of Gaussian mixture model,so as to improve the Gaussian mixture model and EM algorithm,so as to solve the problem of segmentation instability caused by difficult adjustment of parameters.This chapter also proposes two methods to determine the optimal number of clusters for segmentation,which makes the segmentation results look more reasonable.