植根区域生长法(SRG)中的边界像素及其改进算法

根植区域生长法(Seeded Region Growing)是处理图象分割的一种快捷的半自动算法([1]，R．Adams and L．Bischof，1994)，也是目前处理图象问题的一种公认的有效方法和技术．边界像素对该算法的行为及效果有着微妙的但却是重要的．在应用中我们发现，当分割块处于狭窄连通状态时，这一算法常常是失败的，而失败的结果通常会导致原始图象的细节失真或者过分地依赖种子的选择．针对这一问题，我们对R．Adams和L．Bischof的植根区域生长法进行了改进，给出了一个有效算法，解决了原来算法对狭窄连通分割块失败的问题．

On boundary pixels in seeded region growing segmentation

Boundary pixels can have a subtle but serious impact on the performance of the seeded region growing algorithm[1], a fast semi-automatic algorithm for segmentation of intensity images. We have found, however, that as originally described the algorithm will fail if some sections of a connected segment are narrow. The failure results, for example, in a loss of details or an undue dependence of seed placement. The remedy is a careful processing of the boundary pixels. We here present an algorithm whose performance is not affected by such situations.