基于Hausdorff距离的多尺度轮廓匹配算法

在提取碎片轮廓线的基础上，提出一种以Hausdorff距离为度量准则的轮廓匹配方法。该方法的基本思想是首先采用卷积积分的方法，在多尺度空间里通过对轮廓进行重采样来计算轮廓上每一点的曲率并选取特征点。特征点及其两侧的若干点构成轮廓的特征段，然后由特征段之间的Hausdorff距离来比较二者的相似性，即为匹配程度。当Hausdorff距离小于给定的容差时，说明与特征段相对应的轮廓是匹配的。实验证明该方法是有效的。

A Multiscale Method of Contour Matching Based on Hausdorff Distance

On the basis of contour extraction of fragments, a novel contour matching algorithm was presented, which makes use of Hausdorff distance as measurement criterion. The main idea behind this method was firstly to utilize convolution integral to calculate pointwise curvature through resampling the contour in multiscale space, and then the feature points were selected. Every feature segment consisted of the feature points and some other points in their neighborhood. The Hausdorff distance between feature segments indicated their similarity, namely, the match degree. If the Hausdorff distance is less than the given tolerance, the corresponding contours are suited. The method proves to be effective by realistic experiments.