| Segmentation and labeling of 3D shapes into meaningful parts is fundamental to shape understanding and processing. This problem has wide application like geometric modeling, manufacturing, animation and texturing of 3D meshes in which tasks the segmentation into parts is required. We employ multi-class SVM (Supporting Vector Machine) that is a commonly used supervised learning algorithm to reach the goal of simultaneous segmentation and labeling of 3D meshes. Using geometric features extracted from the point cloud of labeled meshes as training data. Then we use another layer of classifier, which is a binary classifier, to improve the boundary edges in the meshes. In this process, we just learn the boundary edges with the pairwise features that estimate the relationship between faces. To remove the noise points which are the isolated points to the boundary edges, we apply K-means that is an unsupervised clustering method find the groups of the boundaries and together with their group centers, then find the noisy points that has larger distance with its center compared to the average distance. The experimental results show that this method effectively works and has the comparable performance to state-of-the-art. |
