基于特征的自适应正则化配准算法

摘要

Andriy Myronenko提出了一种自适应正则化的方法并将其应用于非刚性图像的配准,该方法在配准速度和配准精确度方面都取得了比较好的效果;但该方法对变形场初始值比较敏感,选择不当则会陷入局部极小值而不能得到理想的配准结果.为了使原始算法得到更广泛的应用,引入了基于特征点的粗配准方法,得到了与真实变形场更加接近的初始变形场,从而摆脱了局部极小值的困扰,得到了正确的配准结果.实验证明,改进后的算法在应用范围和配准精度上都有了提高.%Andriy Myronenko proposed an adaptive regularization method and applied it to non-rigid image registration. This method could achieve a good effect at registration speed and accuracy. But it made a failure registration because of trapping into local minima when chose the initial deformable filed irrelevantly. To using the algorithm extensively , this paper gained the initial value of the deformable field by rough-registration based on feature point. Because the initial deformable field was close to the real value,it could overcome the local minima and format a correct result. By this way,it enhances the practicality and the accuracy of the algorithm.