基于阈值约束最小生成树算法的区域合并方法

为了解决基于形态学的分水岭分割算法受噪声影响而产生的“过分割”问题,提出了基于阈值约束最小生成树算法的区域合并方法.利用图论中最小生成树算法(prim算法、kruskal算法和boruvka算法),把分割后的区域看作是图的顶点,有相邻关系的区域看作是图的边,相邻区域的特征差异看作是边的权值,通过设置合适的阈值和迭代次数进行区域合并.实验结果表明,该方法保持地物边界的同时能够快速有效地合并“过分割”区域,尤其适用于地物复杂、尺寸较大的遥感图像.

Region merging method based on threshold constrained minimum spanning tree

To solve the problem of over-segmentation caused by noise when applying watershed segmentation based on morpho-logy,a new regions merging method is proposed.Based on the minimum spanning tree algorithm(prim,kruskal,boruvka),segmented region is regarded as the vertex of a graph,adjacent regions are regarded as the edges whose weight is determined by the difference between the segmented region and adjacent region,regions merging is completed by setting the appropriate threshold and iteration times.Experiment indicated that the method can keep the boundary of region while merging"over-segmentation" regions quickly and effectively,and it is effective for remote sensing images of complex and large size.