Abstract: Registration of myocardial perfusion MRI images is a typical method in heart process and analysis.For heart MRI time-series images，an improved sub pixel match approach based on cross-weighted moments and mutual information is presented，which adopts mixed registration strategies.Firstly，the cross-weighted moments are used in rough registration to rid those local extremums far away from a global optimal point.And then mutual information，combined with Powell and SA hybrid optimization algorithms，is taken as objective function to make accuracy and speed of image registration greatly improved.

Key words: sub pixel matching, Powell algorithm, simulated annealing algorithm

摘要： 心肌灌注磁共振图像系列的配准是心脏分析诊断的重要辅助手段。针对心脏磁共振时间序列图像，提出了一种结合力矩主轴法与互信息法的序列图像亚像素配准新方法。给出了一种由粗配准到精配准的混合配准策略。采用力矩主轴法进行粗配准，避免那些远离全局最优点的局部极值;采用Powell算法和模拟退火算法相结合的搜索策略，以互信息的值作为目标函数，对结果进行优化。使得图像配准精度和速度有极大的提高。

关键词: 亚像素匹配, Powell算法, 模拟退火算法