一种基于遗传算法的双T-Snake模型图像分割方法

An Image Segmentation Method of Dual T-Snakes Model Based on the Genetic Algorithm

Snake的初衷是为了进行图像分割,但它对初始位置过于敏感,且不能处理拓扑结构改变的问题。初始位置的敏感性可以用遗传算法来克服,因为它是一种全局优化算法,且有良好的数值稳定性。为了更精确地进行图像分割,本文提出了一种基于遗传算法的双T-Snake模型图像分割方法,它将双T-Snake模型解作为遗传算法的搜索空间,这既继承了T-Snake模型的拓扑改变能力,又加快了遗传算法的收敛速度。由于它利用遗传算法的全局优化性能,克服了Snake轮廓局部极小化的缺陷,从而可得到对目标的更精确的分割。将其应用于左心室MRI图像的分割,取得了较好的效果。

The original purpose of Snakesis is for image segmentation. The method suffers from a strong sensitivity to its initial position and can not deal with topological changes. Its sensitivity to initialization can be overcame by the genetic algorithms (GAs). The GAs is a global optimal searching algorithm and has better numerical stability. But its disadvantages are the computational complexity and the rapid increasing of computation by the augmentation of the search space. They both affect the convergence rate of the GAs. This paper presents an image segmentation method of Dual T Snakes model based on the GAs. By making use of the Dual T Snakes model, it inherits the capability of changing the topology of the T Snake, reduces the valid search space for the GAs to remedy its limitations. The solution of the Dual T Snake consists of two curves enclosing each object boundary, and it is composed the valid search space of the GAs. The optimal object boundary can be obtained through the operation of selection, crossover, and mutation. The new model can accelerate the convergence rate while inheriting the capability of changing the topology of the T Snake, avoid local minima from Snakes model, and maintain the global optimal ability of the GAs, then obtain more precise segmentation. Better results are achieved in application of this method on segmentation of cardiac magnetic resonance images.