Multi-model image registration has received much attention at present A new multi-model image registration approach by combining remnant image re-treatment with corner point detection is proposed in the paper. Firsdy, it is found that the so-called transformation parameters in original images can be retrieved by solving those in the remnant images. This fact in turn demonstrates feasibility that bi-dimensional local wave decomposition can also be employed in the multi-model image registration. Secondly, a new high-accuracy corner point matching algorithm based on the Harris corner point detection is described in this paper, while fully making use of graphic point features, gray-level pixel and location information; in particular, a circular template is adopted for calculating similarity indices and implementing relaxation matching. Finally, experiment results show that the algorithm results in automatic matching and it is superior to most existing point matching algorithms in term of accuracy.%目前多模态图像配准研究是现代图像处理的一个热点.提出了一种基于二维局域波分解后剩余图像和角点检测的自动点匹配的多模态图像配准方法.首先通过理论分析得出通过求解剩余图像之间的变换参数即可获得原始图像之间的变换参数,从而论证了二维局域波应用于多模态图像配准的可行性,提出了一个新的基于Harris角点检测的高精度自动点匹配算法.该算法充分利用了图像的角点特征以及图像灰度和位置信息,匹配过程中采用圆形模板进行相似度计算和松弛匹配.实验表明,该算法可以精确的自动匹配,其精度优于传统的点匹配算法.
展开▼
