分水岭变换是数学形态学的主要分割工具，它通过对梯度图像进行分割，能够提供单像素宽的封闭的区域边缘。但是，直接对梯度图像进行分水岭变换存在严重的过分割问题，这个问题往往采用基于标记的分水岭变换加以改进，可是，有时又会带来欠分割问题。通常过分割和欠分割问题主要发生在灰度的非极大值或极小值的过渡区域。为了克服分水岭变换易造成过分割的问题，通过对分水岭变换产生的过分割以及标记点选取可能导致的欠分割原因进行分析，提出了一种基于重构的分水岭变换算法，该算法首先通过基于重构的变换序列滤波来平滑图像，以减少目标内部差异；然后针对梯度图像，利用重构闭来消除灰度过渡区域在梯度图像上所造成的极小；最后对其进行分水岭变换，可有效控制分水岭变换的过分割现象。

Watershed transform is the main tool of mathematical morphology used for image segmentation which can produce one-pixel wide and close edge.This method applied directly to gradient image causes severe over-segmentation.In this paper，the author presents a multi-scale watershed transform based on morphological scale-space by the analysis to the reasons of over-segmentation produced by watershed and the possible under-segmentation due to watershed-plus-marker.The method proposed here could be divided into three steps: (1)The alternate sequential filters(ASF) based on reconstruction are used to reduce some difference inside objects； (2)The closing by reconstruction be applied to gradient image in order to remove some local minima that does not correspond to the maximum and the minimum of the original image； (3)The processed gradient image is segmented using the watershed transform.