中文摘要：

提出了一种基于空间关系一致性的图像自动配准方法，处理具有全局仿射变换的图像配准问题。首先利用Mean-shift算法和区域标记算法从参考图像和待配准图像中提取区域特征作为配准基元；然后采用高斯函数平滑后的区域特征重心点坐标之间的欧式距离描述匹配特征的空间关系一致性，从而定义一个以匹配矩阵为自变量的空间关系一致性度量函数，使它在两幅图像的最佳配准位置达到最大值，将图像配准问题转化为一个最优化问题；最后利用一个局部迭代优化算法求解。高斯函数的方差是度量函数中惟一的预设参数，但它对优化求解的结果基本没有影响，在实际中容易设定，保证了方法的自动性、稳健性和适应性。实际图像的实验结果表明本文方法能有效解决具有全局仿射变换的图像配准问题。

英文摘要：

An automatic image registration method based on spatial relation consistency is pro- posed to deal with the registration of images with affine geometric distortion. Firstly, regions extracted from the reference and sensed images by using the mean-shift method, and the region labeling algorithm are regarded as registration elements; Secondly, the Euclidean distance of the gravities of the matched region pairs filtered by the Gauss function is used to describe the spatial relation consistency of their space position, then a spatial relation consistency measure-ment function of the match matrix, which is the independent variable, is built and its global maximum is assumed to be reached when the sensed image is completely aligned with the reference image; Finally, a local iterative optimal algorithm is used to solve the optimal problem. The variance of the Gauss function is the exclusive preset parameter in the new method and affects the optimal result slightly, thus ensuring the method to be robust and adaptive. Experimental results of real images prove that the proposed method can solve the registration problem of images with affine geometric distortion.