Three-dimensional Model Segmentation Based On Weak Convexity And Shape Diameter Function

The 3D scattered point cloud segmentation model as the basic steps of 3D retrieval,texture mapping,animation and geometric deformation and simplified model applications,widely used in the field of virtual reality,medical diagnosis,product design,3D games etc..Now,unsupervised and no manual input of any parameters can be applied to automatic segmentation of various types of models.To solve this problem,a method of unsupervised and full-automatic 3D scattered point cloud segmentation is proposed based on weak convexity and shape diameter function.It solves the problem of over segmentation and under segmentation,and establishes active contour model to achieve smooth segmentation.In this paper,based on the concavity and convexity and the weak convexity,the spectral clustering method is proposed to realize the over segmentation of the scattered point cloud model.In order to effectively express the similarity between the scattered point cloud data point,this paper presents a k nearest neighbor search method to solve the scattered point cloud point density uneven distribution problem,improve the search accuracy of k neighbor points,each point estimation method to estimate the Mahalanobis distance by MM and robust combination method to estimate the bump and K nearest neighbor points between the combination of the establishment of similar matrix,said topological relation in the point cloud data of each point and its K nearest neighbor points and similarity relation.On the basis of the similar matrix,the Laplasse matrix is obtained by matrix transformation,and its eigenvalues and eigenvectors are normalized.Based on the eigenvectors corresponding to the former K eigenvalues,an improved K-means clustering algorithm is used to realize the over segmentation of the scattered point cloud model.Then,a method of calculating the shape diameter function on the three-dimensional point cloud model is proposed.The concave and convex degree of the boundary is calculated by the normal direction of the point,and the method of merging the over cut block is done through the shape diameter function and the boundary concavo convex degree.In order to solve the problems such as depression and serration on the boundary produced by the above segmentation methods,a new active contour model which can be applied directly to the scattered point cloud model is established to smooth the segmentation boundary.The segmentation algorithm proposed in this paper is implemented through MATLAB and C++ language,and the segmentation results of all kinds of models in two datasets are obtained by semantic segmentation.The experimental results are quantitatively analyzed by four criteria of segmentation evaluation,and compared with the other four unsupervised methods.The experimental results show that the proposed hybrid segmentation method outperforms the other four unsupervised methods.