Animated Character Model Segmentation And Labeling Method Based On Semantic Rule

Animation industry is an important branch of culture industry, whereas the modelingprocess of3D animated character, which is the soul of animation industry, is still inefficientand complex. The development of computer technology makes model retrieval andgeometric modeling possible. Whereas fast modeling of animated characters requires modelsegmentation and labeling results.3D mesh segmentation into meaningful parts isfundamental to model processing. Numerous tasks in geometric modeling, model retrieval,mesh editing, animation and texturing of3D meshes rely on their segmentation in parts, andmany of these problems further require labeled segmentations. There are many efficientapproaches to dealing with different circumstances, however only a few methods can labelmeshes by example with different segmentation type. This thesis proposes a method ofmodel segmentation and labeling for animation character under the guidance of semanticrules, through which user can acquire expect labeled segmentations.The main works of this thesis are as follows:1) Firstly, the thesis proposes description of animation character semantic rule. Thesemantic rule, combining with user’s intent, provides the principle of component labelingas the guidance in the segmentation process. It can reduce human-computer interaction andalso has good adaptability; A friendly tool is designed for rule creation, in order todetermine segmentation granularity and semantic information rapidly.2) Secondly, some geometric features of meshes need to be calculated in thesegmentation process. The more efficient algorithm we use, the better result will be gain.This thesis expounds the principle of mesh geodesic and vertex curvature, and provides adetail description of Dijkstra shortest path algorithm, geodesic algorithm presented byMitchell, Mount, and Papadimitriou, as well as Meyer’s algorithm for calculatingcurvature.3) Finally, a system is designed and implemented for mesh segmentation and labeling.Analyzing semantic rule, detecting end vertexes, dividing by curvature, and obtainingsegment boundaries by region growing method; the boundaries are optimized using shortestproximate geodesic method in their neighbor meshes; the semantic information of eachsub-component are labeled to segmentation results in terms of the guidance of modelsemantic rule.