Research On Shape-Based Feature Extraction Method Of 3D Model

With the development of three-dimensional modeling techniques and computer hardware/software technologies, the application of 3D models is more and more widely and the number of 3D models is growing larger and larger. It can greatly reduce the workload of the design of the new models by making full use of the existing 3D data resources. Content-based 3D model retrieval technology has broken through the limitations of traditional keyword search and becomes a research spot recently.The properties of the 3D model are shape, color and texture. However, color and texture are uncertain in practice. Shape-based 3D model retrieval is the main research content. There are two categories of 3D model feature: global feature and local feature. Two 3D model feature extraction methods are proposed in this paper.The principle plane is a reference plane of 3D model. It satisfies that the sum of distances form the vertices of the 3D model surface to this plane is minimal. So it can be used as a projection plane. According to the experiments, the projection images of 3D models in the same class are similar. The feature of the projection images can be used in the 3D model retrieval. In experiments, the Euclidean distance is used to compute the similarity. According to the P-R plot of the retrieval results, the principle descriptor exceeds the shape distribution.The cone curvature is a good description of local shape. Geodesic distance wave is used to compute cone curvature in this paper. According to the definition of cone curvature, the module of the cone curvature in sharp area is bigger and the one in flat area is very small. So k-means clustering algorithm is used to cluster the cone curvature. The cluster with the biggest module is defined as the salient local feature and the vertices where these salient feature lie are defined as the salient vertex. According to the experiments, the salient vertices of 3D models computed by our method are effective and steady. Based on the PSB models, using EMD distance to compute model similarity, the salient features are tested in the 3D model retrieval. The retrieval results prove that the salient features have the ability of recognizing 3D objects.