融合边缘特征与区域特征的高分辨率遥感图像分割方法

Segmentation of high-resolution remote sensing images based on fusing edge and region features

利用图像的边缘特征和区域特征，设计了一种能实现分割对象边缘精确、对象内部一致性好的分割效果的高分辨率遥感图像分割方法。首先采用canny方法获得边缘，并采用形态学方法改善边缘平滑与间断问题；再利用mean shift方法对图像进行边缘保持的滤波操作；最后，融合边缘特征与区域特征，利用mean shift方法实现图像分割。通过对高分辨率遥感图像进行试验，得到了边缘精确且平滑的区域内部更加一致的分割结果，且在很大程度上消除了欠分割和过分割情况。

To obtain a better result of segmenting high-resolution remote sensing imagery, the paper presents a useful seg- mentation method to achieve segmentation objects with high edge precision and good internal consistency by use of edge and regional features. It is briefly described below： Firstly, the smooth edges are obtained by the canny algo- rithm and the edge gaps are connected with morphologic methods. Secondly, the image is filtered by a mean shift algorithm, which is a kind of edge-preserving filter. Finally, the fusion of the edge and regional features is adopted by the mean shift algorithm to segment the remote sensing image and obtain the segmentation objects which have the smooth edges and uniform regions. The results of the contrastive experiments on a high-resolution image show that the mean shift segmentation by adopting the edge and spectral features is better than the method adopting only the spectral feature.