Abstract: Mutual information-based image registration has been widely exploited due to its high-level automation and accuracy, especially in medical image registration. However, the object function suffers from the local extremum induced by interpolation, which obstructs optimization algorithms and leads to the wrong result of registration. The mechanisms how two common used algorithms work and the reason why the artifacts occurred are analyzed and a new interpolation algorithm is proposed in this paper. Experimental results indicate that the new interpolation algorithm can effectively avoid the local extremum, and smooth the object function through mutual information.

摘要： 基于互信息的图像配准方法具有自动化程度高、配准精度高等优点，近年来在医学图像配准中得到广泛应用。但是当变换后像素坐标位于非采样网格点时，插值算法有时会使目标函数产生局部极值，使得最优化搜索终止于局部极值，得到错误的配准结果。本文分析了两种常见的插值算法和产生局部极值的原因，在此基础上提出一种新的插值算法。实验结果表明，该算法有效地抑制了基于互信息的目标函数的局部极值问题，使目标函数更加平滑。