改进的Powell医学图像配准算法的应用

摘要

The medical image registration method based on mutual information similarity measure and Powell optimization algorithm is slow and easy to fall into local minima. To solve these problems, the calculations of joint histogram and the formulas of normalized mutual information were improved. The improved method simplified the calculation process and the similarity measure could reflect the registration degree more accurately. Moreover, to solve the problem of excessive relying on initial point in Powell optimization algorithm, an improved algorithm was designed to obtain the initial registration parameters, which could avoid the local extremum and reduce the number of iterations. Experimental results show that the improved method for medical image registration can achieve sub-pixel accuracy, increase arithmetic speed obviously and is more stable than ordinary Powell algorithm.%以互信息为相似性测度的Powell优化算法,在图像配准时存在速度慢、容易陷入局部极值等缺点.本文提出了改进的归一化互信息的计算公式及其联合直方图的计算方法,使得在计算过程简化的同时,相似性测度能更准确地反映配准程度.另外,针对Powell优化算法对初始点过度依赖的问题,本文设计了一种改进的Powell算法来求取配准的初始变换参数,能有效避免Powell陷入局部极值,同时减少了迭代的次数.实验结果表明,用改进的方法进行医学图像配准可以达到亚像素精度,在速度上有了明显的提高,比传统的Powell算法更稳定.