水平集理论在磁共振脑图像分割中的模型研究

灰度不均匀性常出现在医学图像中,给图像分割问题带来很大困扰.为了提高鲁棒性,可在分割模型中引入各种先验知识,例如形状和灰度分布信息.而传统的引入先验知识的分割算法,如神经网络算法,仍存在许多问题,包括数据计算量大和边界不连续等.为了解决这些问题,提出了一种基于水平集理论的分割算法.利用局部区域的灰度信息定义能量函数,然后根据能量函数的最小化机制引导水平集曲线进化并最终收敛到目标边界.在仿真实验中,将局域化的水平集算法与传统的自组织映射神经网络算法进行比较.结果表明,所得到的算法在鲁棒去噪和目标边界的连续性方面效果更佳.

Survey on Model of Level Set in MR Brain Image Segmentation

To solve the problem of uneven gray in medical image segmentation, a level set based image segmentation algorithm is proposed, which applies intensity information in local regions defining energy function, and then level set curve can evolve to object boundary according to energy minimization scheme. In simulation, the local level set algorithm is compared with the traditional SOM neural network algorithm. Results show that the novel algorithm can robustly handle noise, and capture continuous obiect boundaries.