Study Of Segmentation Algorithm For Three-dimentional Point Cloud Data Of Blood Vessels Based On Spectral Graph

Real and clear vascular image plays an important role in the process of the diagnosis and treatment of the cerebrovascular disease.Compared with traditional two-dimensional vascular images, 3d vascular images had been used more and more. We can get the 3d point cloud data through the image of the blood vessels obtained from the 3d scanning instrument after further processing. The original three-dimensional vascular branch point cloud data does not have any information, so that it is need to be split for the subsequent applications such as 3d reconstruction. Clustering algorithm is an important method of the point cloud segmentation, one of the emerging spectral clustering algorithm with complex topology point cloud clustering ability. This paper plans to study segmentation algorithm based on spectral clustering for three-dimensional vascular point cloud data.The study has set up the topological relations between original isolated point cloud based on the spectral graph theory and the method of k neighbor. This paper proposes a new method of spectra based on neighbor adaptive scale coefficient matrix to measure the closeness of relationship between point cloud data point effectively. The eigenvalues and eigenvectors of Laplacian matrix of point cloud carry the important information needed for segmentation. We establish the figure matrix on the basis of Laplacian matrix and analyze its properties. For the dealing with of the eigenvalue and eigenvector, we study two methods of spectral clustering based on the second smallest eigenvalue and based on the former k eigenvalues and eigenvectors With the introduction of p-Laplacian operator, the standard spectral clustering boils down to a special case of p = 2 and we establish a method of spectral clustering based on p-Laplacian operator.We write the program of spectral clustering algorithm in this paper based on C + + language and platform of Linux system. We carry on the contrast experiment for part of the blood vessels of the point cloud data based on the new spectra matrix method we proposed with the traditional spectrum matrix method and illustrate the efficiency of new method For complete blood vessels of point cloud data, contrast experiment was carried out based on three kinds of spectral clustering algorithm. And by analyzing the image segmentation, we obtain the advantages and disadvantages of different spectral clustering algorithms, and analyze the influence of sampling radius for segmentation and the time complexity of algorithm.