Chan-Vese图像分割模型的快速实现算法的研究

重初始化是水平集方法耗时的一个主要原因,通过将水平集函数与符号距离函数的偏差作为能量项引入C-V模型,以此来约束水平集函数成为符号距离函数,从而去掉了重初始化步骤。在检测多目标时,采用了曲线族代替单曲线作为初始曲线。在数值算法上,利用加性可操作分裂算子(AOS)消除了对时间步长的限制,可以选择大步长,从而加速了演化过程而且绝对稳定。实验结果表明,分割速度相对于经典的C-V模型有了很大的提高,而精度损失可以忽略。

Research of Fast Approach for Chan-Vese Model

The reinitialization process is quite time-consuming in level set method. A new variational formulation which is the difference between the level set function and the signed distance function is introduced to C-V model. In this way, the level set function is forced close to a signed distance function, and therefore completely eliminates the need of the constly re-initialization procedure. Also, when detecting more than one object, a group of curves are choosed to be the initial curve. In the numerical implementation, AOS scheme is adopted to eliminates limit of time step. So the large time step can be selected to accelerate the evolution velocity. The result shows that the evolution velocity is greatly improved and the precision doesn't reduce.