基于随机轮廓匹配的快速图像配准算法

为了克服确定性图像配准算法计算速度和准确率难以同时兼顾的缺点,提出了一种对部分重叠的图像进行快速配准的方法,该方法是基于轮廓特征的随机匹配算法。通过提取轮廓上的“关键点”作为特征点,随机选择若干特征点对得到候选变换,随后的投票阶段对其变换参数进行检验和求精。实验结果表明:对于典型的应用,该算法比传统确定性匹配算法的速度提高了约一个数量级,能够在线性时间内完成对两幅图像的配准,而且,该方法能够对包含相当比例误报的特征点集进行匹配,具有很强的适应性。

Fast image registration algorithm based on randomized contour matching

An algorithm was developed for fast planar registration of partly overlapping images to overcome the tradeoff difficulty relating to traditional deterministic image registration algorithms.The algorithm uses randomized contour matching,which extracts the "critical points" of the contours as feature points and then randomly selects a subset of feature points to instantiate a candidate transform,followed by a voting process to validate and refine the transform parameters.Experimental results show that the algorithm can match two images in linear time,which is about one magnitude faster than traditional algorithms in typical applications.The algorithm is capable of matching feature points containing a significant percentage of gross errors,which guarantees the matching algorithm to be robust.