基于分水岭变换的多尺度遥感图像分割算法

Multi-scale Image Segmentation of Remote Sensing Image Based on Watershed Transformation

分水岭变换是一种适用于图像分割的强有力的形态工具,能够自动生成一系列封闭分割区域。分水岭变换的不足之处在于它的过分割结果。为了克服分水岭变换固有的过度分割现象,利用非线性滤波和改进的快速区域合并算法优化分水岭变换得出的初始分割结果,并针对高分辨遥感图像所体现出来的地物的多种信息特征,结合多种特征进行了区域合并。实验结果与MeanShift算法得到的结果进行了比较,证明该算法不仅能充分利用高分辨率遥感图像中地物的信息特征获得良好的分割效果,而且大大减少了计算时间。

Watershed transformation is a powerful morphological tool for image segmentation, which can automatically generate a series of closed segmentation regions. However, the watershed transformation might give rise to over-segmentation. In order to overcome the inherent drawback of watershed algorithm-over-segmentation, a kind of non-linear filter algorithm and a modified fast region merging algorithm are proposed to improve initial segmentation result obtained by the watershed transformation. According to the all kinds of characteristic explored from the high resolution remote sensing image, the modified fast region merging is implemented combined with multi-characteristic. Experimental results show the proposed algorithm not only can obtain the good segmentation result by making good use of the characteristics explored from the high resolution image, but also the processing time are drastically reduced.