摘要
利用遗传算法全局随机搜索的特点,可以解决模糊C均值聚类(FCM)算法在医学图像分割中容易陷入局部最优解的问题,但确定遗传算法的初始搜索范围时,需要借助于人的经验。为此,用收敛速度快的硬聚类算法得到的聚类中心作为参考,上下浮动划出一个较小的数据范围,作为遗传算法的初始搜索空间。该方法在避免FCM算法陷入局部最优化的同时,也加速了遗传算法的收敛过程。实验表明,该方法相对于标准的遗传模糊算法,效果要好得多。
The fuzzy C-means algorithm has the limitation of converging to the local infinitesimal point in medical image segmentation, which is optimized by genetic algorithm with the feature of global random search. But the initial searching scope of genetic algorithm is often confirmed by human experiment. A method to select a small data scope as the initial searching space for genetic algorithm is presented according to the clustering centers obtained by the K-means clustering algorithm, While avoiding being trapped in a local optimum, this method speeds up the convergence of genetic algorithm. Test with brain magnetic resonance images shows that the method is better than that of genetic algorithm fused with fuzzy C-means clustering.
出处
《计算机工程与设计》
CSCD
北大核心
2006年第13期2357-2359,共3页
Computer Engineering and Design
基金
甘肃省自然科学基金项目(32S042-B25-014)