结合光谱、纹理与形状结构信息的遥感影像分割方法

On Combining Spectral,Textural and Shape Features for Remote Sensing Image Segmentation

针对目前遥感影像分割中多特征利用的问题,提出一种综合利用光谱、纹理与形状信息的分割方法。该方法在进行初始分割的基础上,统计区域的光谱和LBP纹理特征;然后依据光谱、纹理与形状特征计算相邻区域之间的异质性,并以此为基础构建区域邻接图(region adjacency graph,RAG);最后在邻接图的基础上采用逐步迭代优化算法进行区域合并获取最终分割结果。采用QuickBird和SAR影像的分割试验,证明该算法能充分利用影像中地物的光谱、纹理与形状信息,分割效果良好,效率高。

To utilize multiple features for segmentation of remote sensing images,a novel segmentation method combining spectral,textural and shape features is presented.The method starts with an initial partition,where the spectral and textural features of each region are extracted.And then,the regions are regarded as node set of the region adjacency graph（RAG）,and the weights of the edges correspond to merging cost of adjacency regions.Finally,a hierarchy stepwise optimized region merging procedure is adopted based on the region adjacency graph to get the final result.Experiments on QuickBird and SAR images have shown the effectiveness and high efficiency of the method.