中文摘要：

本文提出了一种改进的基于不变描述子的图像自动配准方法来处理具有全局仿射变换的图像配准问题。该方法主要分为三步：首先,利用Mean-shift算法和区域标记算法分别从参考图像和待配准图像中提取区域特征作为配准基元,采用7个仿射不变矩作为不变描述子 其次,通过组合仿射不变矩的最小距离准则和行、列匹配概率系数算法获得一个候选匹配区域特征集,运用穷举策略从中得到三个最好的匹配区域特征,利用它们的重心点坐标计算仿射变换模型参数的初始值 最后,根据空间一致性检测获得所有正确的匹配区域特征,利用它们的重心点坐标和最小二乘算法估计最优的仿射变换模型参数。仿真和实际图像数据的配准实验和对比实验结果表明本文方法具有较高的可靠性和配准精度。

英文摘要：

An improved automated image registration method based on invariant descriptors is proposed in this paper to deal with registration of images with affine geometric distortion. The method mainly consists of three steps：first, regions are extracted from the ref- erence image and sensed image respectively using the Mean-shift method and region labeling algorithm to be regarded as registration ele- ments and seven affine invariant moments arc considered as invariant descriptors;Second, a candidate matching region pairs set is ob- tained by combining the criterion of the nearest distant of affine invariant moments of regions and row and column matching likelihood coefficients algorithms, and then three best matching region pairs are acquired via exhaustive search and the initial parameters of affine transformation model is computed by coordinates of centers of gravity of them;Finally, obtain all matched regions according to spatial consistency criterion and then the least square method is used to estimate the optimal parameters of affine transformation model by coor- dinates of centers of gravity of them. The experiment results of synthetical and real images prove that the proposed method is comparatively robust and has high registration accuracy.