Abstract: Graphical models have good performance in point pattern matching. However, the method has high computation complexity and attends to be affected by outliers in separators. In order to match the point pattern accurately and efficiently, this paper proposes a coarse-to-fine matching algorithm. A coarse matching process is completed using normal cross-correlation algorithm with windows including feature points, which reduces the number of outliers and improves the matching efficiency. A novel graphical model is proposed. The model can make use of positional information of feature points and gray information of the windows including the feature points. A stepwise matching method is applied to the point pairs matched by normal cross-correlation method and the fine matching result is obtained. The matching experiment results show the proposed method can reduce significantly the running time and improve the matching accuracy.

Key words: graphical model, point pattern matching, normal cross-correlation

摘要： 点模式匹配的概率图模型具有很好的匹配精度，但是计算复杂度较高，当隔离子中包含异常点（outlier）时匹配精度会受到较大的影响。为了提高匹配的速度和精度，提出了一种由粗到精的图模型点模式匹配算法。利用包含特征点的窗口，用标准化互相关方法对特征点进行粗匹配，以减少异常点的数量，提高后续匹配方法的速度和精度。提出了一种新的点模式匹配的概率图模型，这种图模型能综合利用特征点的位置信息和包含特征点的邻域的灰度信息。利用提出的概率图匹配方法对粗匹配所得到的点对进行分段匹配，得到精确的匹配结果。对光学图像和遥感图像的匹配实验显示该方法能显著减少点模式匹配时间，提高匹配的精度。

关键词: 图模型, 点模式匹配, 标准化互相关