图像配准是将同一场景的不同图像对齐或进行广义的匹配,其中基于特征点的方法是应用最广泛的一类方法,而单 一的特征点提取和匹配算法难以实现精确的匹配。文中提出了一种基于互相关双向匹配与匹配支持度相结合的高精度 匹配方法,利用互相关法进行双向匹配得到粗匹配点对;依据初始候选匹配点对的匹配支持度对其排序,选取支持度位于 前面的候选匹配点对,即可排除误匹配点对,优选出真实的匹配点对。采用粗精两次匹配建立特征点之间的对应关系,相 对于传统的单向匹配法有较高的可靠性和有效性。实验表明,这种匹配方法对于旋转、平移、缩放等多种几何变换的情形 都是十分有效的。

Image registration is alignment of different images in the same scene or matchment in generalization. The method based on feature-point is one of the most widely used methods, but simply applying feature-point extraction and matchment algorithm can hardly achieve accurate matchment. Therefore, a high precision method based on bidirectional cross-correlation and matching support level is presented in this paper, in which rough-precise feature points are produced via bidirectional cross-correlation matchment. Then those points are sorted based on different matching support levels. So true and effective feature points are selected and meanwhile false ones are deleted. This method is more reliable and effective than traditional unilateral matchment. Experiments show that this method is applicable for geometry transformation,such as rotation, scale, translation.