对数极坐标变换域下互信息图像配准方法

当待配准中的图像中存在较大的尺度变换时，传统的图像互信息配准方法对图像尺度变换参数很敏感，易于在优化搜索时陷入局部极值点，配准效果不好。文中提出基于对数极坐标（log—polar）域的图像配准方法，它较好地分离了图像缩放参数和图像旋转参数的相互影响。仿真结果表明，该方法在大尺度图像配准场合应用时比通常的互信息方法有更高的精度和鲁棒性。

Mutual information image registration based on log-polar transform

For the 2-D image registration problem, the traditional mutual information metric is not vey effective when the transform scale is very large. The mutual information image registration method is more likely to fall into a local minimum point since the registration precision is sensitive to the transform parameters. To solve this problem, this paper describes a new image registration technique based on the log-polar transform, and it can separate the interaction between the scale and the rotation parameters more easily than the traditional way. Experiment results show our method is more accurate and robust than the traditional MI method for the large scale image registration.