基于一种改进的Powell算法和互信息的医学图像配准方法  

Medical Image Registration Method Based on an Improved Powell Algorithm and Mutual Information

作　　者：刘林[1] 黄英[1] 贺振华[1] 

机构地区：[1]广东工业大学自动化学院,广州510006

出　　处：《电脑编程技巧与维护》2013年第2期78-80,106,共4页Computer Programming Skills & Maintenance

摘　　要：Powell是一种直接法,不用计算目标函数的梯度,仅通过比较目标函数的数值大小来移动迭代点就可求出极值。但Powell算法对参数的初始值有很大的依赖性,在图像配准的优化过程中易陷入局部最优,使得优化结果很大程度上依赖于初始值,会得到错误的配准参数,从而影响配准效果。为解决这一问题,使用粒子群优化算法(PSO)求取Powell算法的初始值。经检验,此方法克服了Powell算法的缺点,大大提高了配准精度。Powell is a direct method, do not calculate the gradient of the objective function only by the magnitude of the comparison of the target function to move the iteration point can be obtained extremum. But Powell algorithm is greatly dependent of the initial value of the parameter, image registration optimization process is easy to fall into local optimum, making the optimization results largely depends on the initial value will get the wrong registration parameters, thus affecting the alignment effect. To solve this problem, particle swarm optimization （PSO） to strike a Powell algorithm the initial value test- ingshow, this method overcomes the shortcominzs of the Powell algorithm, greatly improve the accuracy of the registration.

关 键 词：POWELL算法 PSO 互信息 配准 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]