Search

To improve the efficiency of building the Delaunay triangulation mesh of point cloud data, this paper proposed a new Delaunay triangulation method based on parametric method. Firstly, this paper presented a point cloud segmentation algorithm based on region growing method, ensuring there was no overlap point after parameterization. Then it used a new parameterization method which could remain the distance between points furthest to project three-dimensional point cloud to two-dimensional plane. Thirdly, it reconstructed the point cloud' s Delaunay triangulation in two-dimensional plane and then returned the results to the three-dimensional space. Experimental results show that this method produces triangular mesh with good quality. Compared the experimental results with other algorithms, this algorithm is more efficient because it builds triangular meshes in lowdimensional space. [ABSTRACT FROM AUTHOR]

SVM algorithm has great advantages in dealing with high dimensional data, but it does not consider the problem of semantic similarity measurement. However, LDA topic model can solve the problem of similarity measurement and single theme. In order to get more precise classification and make use of the advantages of SVM and LDA, this paper proposed a new efficient method. Firstly, it studied on LDA topic model for modeling and feature selection in order to determine the number of hidden topic number and topic-document matrix. Secondly, it proposed a new method for calculating the weights which considered the features and categories of correlation based on classical weight calculation. Finally, it applied this new method to the following SVM classifier (wSVM). The experiments were based on R software for categorization of the data obtained from So-gou laboratory. The experimental results are with a high accuracy of macro_Fl 0.943. It verifies LDA-wSVM model has superiority in text categorization. [ABSTRACT FROM AUTHOR]