具有强抗噪声能力的图像分割方法

Method of image segmentation with high anti-noise performance

尽可能多地综合利用图像整体和细部信息 ,是精确分割图像、提取特征的关键。流域思想利用了整体信息 ,应用于梯度图像上 ,兼顾了细部信息。引入待分割物体全局连通性 ,将这种方法用于物体与背景的分割 ,得到了极好的分割效果 ,对噪声有极强的抑制能力 ,同时具有原理简单、分割结果为连续单像素边缘、边缘定位准确、可以同时标记多个区域等优点。给出了形态学流域算法的基本原理及数学实现 ,同时也总结了方法的缺点和解决途径。

The key problem of accurate image segmentation and character extraction is to utilize information of both whole image and local sections as much as possible. Idea of drainage area based on whole image information takes account of local information when it is applied on gradient image. Excellent segmentation result and the ability of restraining image noise are obtained when this approach is used in the segmentation of object and background and the overall connectivity of object to be segmented is lead in. Advantages, such as simple principle, connected single pixel edge output, accurate edge orientation and simultaneous output of different signaled areas, are proven. The principle and mathematical description of the watershed algorithm are shown, and the disadvantages and resolving strategy are summarized.