摘要
基于互信息的配准方法广泛应用于医学图像的配准,但当两幅待匹配图像重叠部分较少或图像包含的信息不够充分时,配准目标函数会出现较多的局部最优解而影响匹配精度。本文构造了新的配准目标函数,在互信息的基础上引入图像梯度信息,使两者相互补充,减少了目标函数中局部最优解的数目,突出了全局最优解。本文还设计了一种新的基于单纯形的模拟退火优化算法,该算法能以较大的概率迅速搜索到目标函数的全局最优解,以获得较满意的配准结果。本文使用该方法对二维医学图像进行配准,实验结果表明,该方法配准速度较快、精度较高,是一种有效的配准方法。
Image registration based on mutual information is widely employed in processing medical imagery. Nevertheless, there are often an excessive number of local optimization results for each registration target function when two images are not overlapped with sufficient areas or if there is not enough useful information contained in the images. This will significantly affect the registration result. This paper introduces a new registration function by combining mutual information with gradient information of the images to be registered. The function can reduce the number of local optimums and make the global optimum more prominent. A new simulated annealing optimization based on simplex search is also presented. The new method was used in 2 dimension medical image registrations. It increased the probability of finding the global optimization results. The experiment results confirm that the method features higher accuracy and efficiency.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2006年第9期1141-1146,共6页
Chinese Journal of Scientific Instrument
关键词
图像配准
互信息
图像梯度
模拟退火
image registration mutual information image gradient simulated annealing optimization
