一种基于分水岭变换的图像分割方案

Image segmentation based on watershed transform

针对传统分水岭分割算法对噪声敏感和易于产生过分割问题,提出了一种基于开闭二次重建和非线性处理的分水岭图像分割方案.该方案根据噪声与信号在尺度与幅度分布上的差异,结合形态学开闭运算的特点,对原始图像进行形态开闭预重建,在计算形态梯度之后采用开闭后重建,对梯度进行给定的阈值变换,引入给定尺度等级的非线性分类,在像素连通关系的基础上,研究了一种改进的分水岭标记算法进行分割,并给出了具体实现流程,.仿真实验结果说明,该方案能够抑制传统算法中的过分割,且边缘定位准确.

Aimed at resolving the problems of sensitivity to noise and over-segmentation existing in traditional watershed algorithm, a watershed segmentation scheme based on double opening and closing reconstruction and non-linear processing was proposed. According to the differences of scale and amplitude distribution between noise and signal in the scheme, morphological opening and closing pre-reconstruction of the original image was utilized by integrating with characteristics of morphological opening and closing operation, and morphological opening and closing post-reconstruction was performed after the morphological gradient＇s calculation. After a certain threshold transform was employed on the gradient, a non-linear classification based on certain scale grade was proposed. An improved watershed labelling algorithm was put forward and applied based on pixel＇s connexity, and the concrete implementation flow was given. Simulated experimental results show that the scheme can restrain over-segmentation existing in traditional algorithms, and exactly locate edges.