基于互信息和混合算法的印刷图像配准

Printing Images Registration Based on Mutual Information and Hybrid Algorithm

目的为了提高印刷图像配准的精度,提出一种基于混合搜索算法的图像配准方法。方法首先求取图像的归一化互信息,然后利用GA算法(遗传算法)进行全局搜索,得出粗配准参数;最后,利用Powell算法进行局部寻优,得出精配准参数。结果混合算法的配准结果与只用单一Powell搜索算法或只用单一GA搜索算法相比,在各个几何变换方向上得到了更小的配准误差。结论与GA算法和Powell算法相比,文中建议的混合算法配准精确度更高、速度更快。

The work aims to propose an image registration method based on hybrid search algorithm, so as to improve the accuracy of printing image registration（IR）. Firstly, the normalized mutual information of the image was calculated, and then the global search was done by GA（genetic algorithm） to obtain the rough registration parameters. Finally, the Powell algorithm was used for local optimization to obtain the precise registration parameters. Compared with the single Powell or GA search algorithm, the registration results of hybrid algorithm had smaller registration error in each geometric transformation direction. The proposed hybrid algorithm registration is much more accurate and fabulously faster than GA algorithm and Powell algorithm.