基于加权相邻关系的形状轮廓点匹配

Point Matching for Shapes Based on Weighted Neighborhood Relationships

轮廓点匹配是形状匹配的一种典型方法。在各种形变情况下,形状轮廓点的相邻关系往往比其他全局关系更稳定。本文在保持局部邻居结构的点匹配算法基础上,引入了邻居的权的概念。首先基于点到邻居的距离为每个点的邻居关系赋权,然后结合形状上下文距离把点匹配问题转化为有向属性关系图匹配问题,用松弛迭代法求解。引入邻居关系的权,使匹配不仅保持邻居集的一致性,同时还保持邻居之间的距离相对关系。实验证明,本文方法能够提高匹配效果,加快匹配算法收敛速度。

Shape matching typically formulated as a point matching problem by describing shape contour as a set of points. The neighborhood structure in point sets is often more stable than other global relationships in varied transformations. We improve the point matching approach which preserves the local neighborhood structures by weighting neighborhood relationships. Relationships of a point and its neighbors are weighted by the distance between them. By introducing shape context, the point sets are then formulated to directed Attributed Relational Graphs, which are matched using relaxation labeling approaeh. Weighting neighborhood relationships makes the matching not only keep the coherence between two matched points; neighbors, but also preserve the order of them. Experiment result shows that our approach can improve the matching efficiency and the converging speed.