Abstract: In order to overcome the disadvantage of being sensitive to model surface local trait and noise in present mesh segmentation algorithm, this paper defines a surface metric called volumetric radius that is invariant to the pose of the model and robust to noise. Based on this metric, a hierarchical mesh segmentation algorithm is proposed. It computes Volumetric Radius(VR) for each vertex, fits the distribution of VR values using two-state Gaussian Mixture Model(GMM), uses connected components algorithm to get the separated sub-parts, and employs a graph-cut algorithm to get the meaningful partitioning. Execute recursively aforementioned last three procedures for every subpart to gain hierarchical mesh segmentation. Experiment indicates that the proposed algorithm is efficient and can partition a mesh into meaningful parts, meanwhile, it is invariant to the pose of the model and robust to noise.

Key words: Volumetric Radius(VR), Gaussian Mixture Model(GMM), mesh segmentation, graph cut, Voronoi diagram

摘要： 为解决现有网格分割算法对表面局部特征和噪声敏感的问题，定义一种刻画模型体属性的测度——体半径，该测度具有姿态不变性和噪声鲁棒性。在此基础上，提出一种网格层次分割算法。计算模型顶点的体半径值，采用二状态高斯混合模型对体半径分布拟合，进行连通分量标注，获取模型的各子部件，结合图切分技术得到有视觉意义的分割结果。实验结果表明，该算法可以实现有意义的网格分割，且对物体的姿态变化及噪声具有较好的鲁棒性。

关键词: 体半径, 高斯混合模型, 网格分割, 图切分, Voronoi图

CLC Number: 

• TP393.03