Research On Methods Of Extraction Of Feature Points On 3D Meshes

With the rapid development of computer hardware and graphics,three-dimensional(3D) models are widely used in various fields of life and production. The 3D meshes gradually become a hot topic in computer graphics due to its simple and intuitive way to represent 3D models. Feature points are the most simple feature element of geometry characteristics on 3D meshes. It can not only convey the basic geometric information on the model, and save a lot of time and memory, but also conduct the post practical application such as model identification and matching,surface reconstruction and others.There are a variety of feature points extraction methods on 3D meshes currently, the most effective one is to combine the local geometry of the mesh with the global features. This method can capture the vertices not only locating on salient areas, but also on the area representing the basic characteristics of the entire model. Feature points set is finally obtained and it can effectively exhibit 3D meshes geometric features. In order to efficiently extract feature points on 3D meshes,we have done some research on a new algorithm,local and global characteristics of the model considered, the contents in this paper are as follows:An improved feature points extraction method is proposed based on data gravitation. Data gravitation is a new concept derived from universal gravitation in physics. And it is applied to feature points extraction on 3D meshes. We first normalize Gaussian curvature of every vertex,and add the vertex with the maximum Gaussian curvature to the initial set of feature points; then we determine whether to weigh the normalized Gaussian curvature according to some condition, and use the principle of farthest point sampling to calculate neighbors for every vertex; next we calculate data gravitation values for each vertex according to the Gaussian curvature and its neighbors; finally we add the vertex with the largest data gravitation to the feature points set until enough feature points are extracted.We compare feature points obtained by data gravitation with other algorithms' results then evaluate all results based on benchmark data. We use three statistical measure: false negative error, false positive error and weighted miss error to assess all methods; subsequently we use Wilcoxon signed rank test to test our method, the hypothesis testing turns out that data gravitation can perform well on population as it does on sample. The results show that data gravitation is applicable to a wide range of models and it is an effective feature points extraction algorithm.We also improve a method which combines principal curvature with bilateral filtering algorithm to extract feature points. After using principal curvature as the processing target, bilateral filtering will be applied to 3D vertexes. We select the significant vertices and add them to feature points set. By comparison with other algorithms, the evaluation and hypothesis testing results shows that this algorithm can extract more accurate feature points.