Research On 3D Mesh Segmentation And Skeleton Algorithms Based On Feature Points

Since the mid 20th century, with the improved computer performance and the rapid development of network bandwidth, the fourth generation digital media whose main form is 3D model is being more widely used, as following the audio, image and video. With the 3D model applied delpthly, people are not limited to learn the data itself, but desire to learn from the content. And many 3D mesh algorithms also want to have a synergistic relationship with some advanced 3D mesh algorithms. 3D mesh segmentation and skeleton extraction is that two tools.In this paper, first we will give a improved feature points extraction algorithm, and second we will give a mesh segmentation algorithm based on feature points, and third we will give a mesh skeleton extraction algorithm based on the segmentation components and feature points.Our feature points extraction algorithm works on mesh. First we find the two vertexs whose geodesic distance is farthest as two source points. Second, we define two scalar functions as the distance of each vertex to that two source points in the form of geodesic, and then we can get two key point sets by that two functions. Third, by calculating the generalized intersection of that two key point sets, we can get the feature points of this mesh.Our mesh segmentation algorithm first calculate the two markers of the central component based on feature points, and second import the two markers to the feature points set, and third part the mesh with markov random field and graph cut.Our mesh skeleton extraction algorithm first part the mesh using the above mesh segmentation method, and then extract the skeleton of each segmentation component based on the principal axis and the feature points. But we do not connect the skeletons of all segmentation components.Experimental results show that the improved feature points extraction algorithm is suitable for our segmentation algorithm, and our segmentation method can result in meaningful parts, but unfortunately it sometimes leads to over segmentation. And the result of our mesh skeleton extraction method can reflect the original model topology well.