高斯混合模型改进的活动轮廓模型MRI分割  

MRI Segmentation via Active Contour Model Improved with Gaussian Mixture Model

作　　者：张建伟[1] 夏德深[1] 

机构地区：[1]南京理工大学计算机系

出　　处：《计算机辅助设计与图形学学报》2005年第12期2647-2653,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：香港特区政府研究资助局研究项目(CUHK418001E;CUHK100C)

摘　　要：引入高斯混合模型逼近图像的直方图概率,利用遗传算法的全局优化特性求解高斯混合模型的参数,以提高参数精度;由此构造新的活动轮廓模型约束项.在新的约束项作用下,活动轮廓模型可以有效地减少噪声的影响,防止从弱边界泄漏.对左心室核磁共振图像的分割实验表明,该模型具有较好的分割效果.Gaussian mixture model is used to approach the probability of the image＇s histogram for its veracity. By its feature of global optimization, the genetic algorithm is employed to calculate more accurate parameters of Gaussian mixture model. With these parameters a new sanction can be made to improve active contour models, to reduce the noise effect and prevent the curve over the weak edges. Therefore, the models fit better with MRI. Experiments on the segmentation of left ventricle magnetic resonance images show that this model performs better.

关 键 词：高斯混合模型 EM算法 遗传算法 图像分割 核磁共振图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]