基于改进分水岭算法的高分遥感影像分割方法研究

Segmentation of High Resolution Remote Sensing Image Based on Improved Watershed Transformation

图像分割是遥感图像影像分析的前提基础和关键步骤,也是图像处理的经典难题。分水岭变换则是一种适用于图像分割的强有力的形态学工具,然而其不足之处在于它的过分割结果。提出一种形态学重建和标记分水岭的分割方案,该方案首先使用Sobel梯度重建得到具有边缘信息的梯度图像;然后利用形态学开闭重建对灰度图像进行滤波处理,并计算出二值标记图像;在此基础上再进行基于标记的分水岭变换。实验结果表明,该方法分割精度明显提高,并能够较好地抑制过分割现象。

Image segmentation is based on the premise and key step of remote sensing image analysis, image processing is a classic problem. Watershed transform is a suitable tool for powerful morphological image segmentation, but its weak point lies in its over-segmentation results. Morphological reconstruction and labeling watershed segmentation scheme is proposed and it uses Sobel gradient to reconstruct the gradient image with the edge information, then uses morphological opening and closing reconstruction to filter gray image, and calculates the binary tag image, on this basis, then tag-based watershed transformation is carried out. Experi- mental results show that this method significantly improve segmentation accuracy, and better suppress over-segmentation.