一种新颖的快速水平集初始化方法

A Novel Fast Level Set Initialization Method

水平集方法是偏微分图像分析的一个重要的数值方法,其最大的问题是计算量大,速度缓慢,特别是需要频繁的初始化。针对这个问题进行深入研究,提出了基于数学形态学的水平集初始化方法。形态学方法分为平滑和重建两个步骤,利用基本的开、闭和膨胀运算来分别平滑数值不精确的距离函数和重建窄带。新的方法不仅具有很高的效率,而且可以适应各种拓扑变化,完全可以替代常规方法,运用于数字图像的实时处理中。

Level set method is an important numerical scheme to PDE image analysis, while the most serious problem is that level set is accompanied with high computation cost, slow speed, and especially frequent re-initialization. The paper carefully researches the disadvantages and presents a new initialization method based on mathematical morphology. The new method includes two schemes, i. e. smoothing scheme and reconstruction scheme, which perform two operations of smoothing inaccurate distance function and rebuilding narrow band separately using open, close and expansion operations. The proposed method is not only of high efficiency, but also can deal with the topology variations adaptively. This method can replace the conventional methods successfully and can also be applied in the real-time image processing.