دانلود رایگان مقاله زمینه های شکل حجمی برای تقسیم بندی مش

Abstract

In the field of mesh segmentation, co-segmentation techniques achieve state-of-the-art performance; however, the segmentation results rely on the shape descriptors used in the segmentation process. In this paper, we propose a novel type of descriptor called the “volumetric shape context” (VSC). For each triangle in the mesh, the VSC describes the distribution of the shape's volume relative to the center of the triangle. This descriptor is descriptive, robust, and invariant under rigid transformations, uniform scaling, mirror imaging and model degeneration. We compare the VSC with state-of-the-art descriptors in a supervised mesh segmentation framework, and the results show that the VSC is most frequently selected as the first descriptor and that combining the VSC with other descriptors improves the segmentation results, thereby demonstrating the descriptiveness of the VSC.

5. Conclusions and future work

In this paper, we propose a novel 3D descriptor for mesh co-segmentation named the “volumetric shape context” (VSC), which describes the volume distribution of a mesh. The performance of this descriptor relies on the auxiliary orientations extracted from the mesh; therefore, we also propose an interactive method and an automatic context-based method of estimating the auxiliary orientations. The experimental results prove that the VSC offers state-of-the-art descriptiveness and can significantly improve the results of segmentation and labeling when it is employed in combination with other descriptors, regardless of which auxiliary orientation estimation method is applied. Although the VSC obtained using the automatic context-based auxiliary orientation estimation method has been proven to be descriptive, the context-based method was not meticulously designed. In our future work, we plan to further improve the performance of the context-based method. For example, the symmetry of the models could be considered. At present, we select the best direction from among 6 possible directions; however, we could also design an algorithm to select the best direction from among all possible directions. Moreover, the consistency of the auxiliary orientations is still an open problem, and the VSC would benefit from the proposal of other effective automatic methods for their extraction. The VSC is designed for use in mesh co-segmentation, including supervised, unsupervised and semi-supervised approaches. In our experiments, the VSC was successfully applied in a supervised approach, but the parameters used in our experiments may not be suitable for use in unsupervised or semi-supervised approaches. Therefore, it will be necessary to investigate what values of the parameters are most suitable in these cases.