基于梯度的混合Mumford-Shah模型医学图像分割

Hybrid Mumford-Shah Model for Medical Image Segmentation Based on Gradient

针对C-V法的水平集图像分割法缺少局部控制能力等问题,将基于边缘的几何主动轮廓线模型和基于区域的C-V法两者结合起来,提出了基于梯度的混合Mumford-Shah图像分割模型HMSG。给出了HMSG模型的参数设置准则,在分割的初期加大模型中全局特征项的权值,在分割的后期则加大局部特征项的权值,以提高模型的图像分割能力。对合成图像与医学图像的分割实验结果表明,该方法优于C-V方法对于含有噪声和边缘模糊的非二值图像的分割,能够较为准确地提取图像边界,可以有效提高图像分割整体性能。

The proposed level set method by C-V is failed to control the local feature. In order to eliminate C-V method＇s demerits, a hybrid Mumford-Shah model based on gradient（HMSG） is proposed. HMSG model has the merits of the geometric active contour based on edge and C-V method based on region. In addition, a rule of parameter choice is given to harmonize simultaneously both regional and gradient information in the processing of image segmentation. The rule is to add the weight of global information in the beginning of image segmentation, and to add the weight of local information in the second stage. The experimental results of the synthetic image and MR image segmentation show that it is often challenging to more obtain a reliable segmentation for noise and unclear edges image than the C-V method.