针对图结构在一些非刚性变换下谱特征不稳定等问题,提出一种基于几何关系直方图的图结构信息的描述方法,并在此基础上实现图的顶点匹配。首先通过夹角和距离等一些几何量来描述图的顶点与其他各边的位置关系,然后结合直方图给出一种图顶点的特征描述,最后采用改进的匈牙利算法实现图的匹配。不同于传统的描述方法,该方法是利用图的顶点与边的几何关系的统计特征来描述图的结构信息。其结构简单,信息描述充分。实验结果表明,方法对于一些扰动前后的图的匹配具有较高的匹配准确度。
For graph structure the spectral features become instable in some non-rigid transform and other issues, a description of graph structure information based on geometry relation histogram is proposed in the paper and then the graph matching is implemented based on the histogram information. The algorithm firstly describes the position relation between vertex and edge by the angle and distance. Then a new description of the vertex feature based on the histogram is proposed. Finally, the algorithm implements the graph matching by improved Hungarian algorithm. Distinguishing from traditional approaches, it describes the graph structure by geometrical feature of the graph vertices and edges which is simple and abundant. The experimental results demonstrate that the algorithm performs higher matching accuracy especially in the case of some perturbation.