中文摘要：

针对高分辨率遥感影像提出一种有效的分割方法。首先采用分块方式计算图像复杂度，根据不同的复杂度，使用相应参数的均值偏移算法对图像进行颜色量化，图像越复杂，量化级数越多；在量化结果的基础上构建多形状结构元素，然后使用该结构元素对图像提取模糊形态学梯度，并使用浸没式分水岭变换得到分割结果；最后，使用改进的合并代价函数进行小区域合并而获得最终结果。对QuickBird多光谱影像进行分割实验，结果证明了该方法的有效性。

英文摘要：

This paper proposed an effective segmentation of high-resolution remote sensing image. At first, calculated the image complexity by imago-blocking. With different image complexities, implemented mean-shift algorithm to quantize image color which using corresponding parameters. The more complex the image is, the more the quantization scalar is. Constructed multi-shape structure elements based on the quantization result and fuzzy morphological gradient was got. Used immersion simulation watershed transform to obtain the segmentation result. At last, merged small regions by using improved merging cost functions. Experiments have been implemented to QuickBird multi-spectrum images and the results show good performance of this segmentation method in quality.