基于能量优化的三维网格模型分割方法

3D Mesh Segmentation Based on Energy Optimization

针对现有的三维网格模型分割方法存在过分割或欠分割、分割线锯齿化明显、人工干预多等问题,提出一种基于能量优化和区分度的三维网格模型分割方法.首先提出能量和区分度这2种鲁棒性更强的特征,用于改善分割边界的精度;其次根据能量、区分度及凹凸性寻找满足条件的分割点,根据点的邻接关系得到分割点集,并基于腐蚀算法细化分割点集以得到分割线;最后结合图的广度优先遍历算法及最小能量原则构造出闭合的分割线.此外,为了提高分割线位置的精度及改善锯齿化明显的问题,采用Dijkstra算法思想进行分割线的优化,得到的分割边界更符合人类视觉.对普林斯顿数据集进行实验,并采用普林斯顿基准同7种一般的分割方法进行定量比较,其中最重要的评估指标兰德指数比7种方法平均高0.21,表明该方法可以得到更高精度且更加符合人类视觉的分割结果.

In order to improve the problems of over-segmentation,obvious zigzag segmentation lines,and too much human intervention in existing 3D mesh model segmentation methods,a segmentation method based on energy optimization and distinction was proposed.Firstly,it used two more robust features,i.e.,energy and distinction,to improve the accuracy of the segmentation boundaries.Secondly,based on the energy,distinction,and concavity,the segmentation points were found;by using the adjacency of the points,the segmentation points sets were obtained;and the segmentation lines were obtained by refining the segmentation points sets based on the corrosion algorithm.Finally,closed segmentation lines were constructed based on the breadth-first search algorithm and the minimum energy principle.In addition,a Dijkstra type optimization method was provided to optimize the shape and position of the segmentation lines.Experiments on the Princeton segmentation benchmark were carried out,and comparisons with seven general segmentation methods under the Princeton Shape Benchmark were done.The most important index,called the Rand index,is 0.21 higher than the seven other methods in average,shows that the proposed method can effectively get more meaningful segmentation results.