中文摘要：

由于高空间分辨率遥感影像自身的复杂性，传统的分水岭分割方法难以取得令人满意的效果。本文提出一种改进分水岭变换的高分辨率遥感影像多尺度分割方法，在抑制分水岭过分割现象的同时，还能实现对遥感影像的多尺度分割。该方法充分考虑了高分辨率遥感影像的多光谱和多尺度特性，首先，利用各向异性扩散滤波技术对影像进行平滑滤波，目的是在滤除各种噪声的同时还能保持影像的边缘特征和重要的细节信息；然后，提取影像的多尺度形态学梯度，并从梯度图像中提取标记；接着进行基于标记的分水岭变换；最后，利用改进的快速区域合并算法实现对影像的多尺度分割。实验表明，改进的算法能有效地抑制分水岭的过分割现象，对高分辨率遥感影像有较好的分割性能。

英文摘要：

With the development of high resolution remote sensing images, imaging analysis technology of ob-ject-oriented method shows a distinct advantage in the field of information extraction and target recognition. Im-age segmentation, as a key technology of object-oriented image analysis method, has a vital role to play on the latter feature extraction and application analysis. Watershed transformation is usually adopted for image segmen-tation because of its unique advantages. However, because of the complexities of high spatial resolution remote sensing image itself, the traditional method of watershed segmentation is difficult to obtain satisfactory results. This paper presents a new multi-scale segmentation method for high resolution remote sensing image based on improved watershed transformation, in order to suppress over-segmentation of watershed transformation, as well as to provide arbitrary-scale segmentation of remote sensing image for object-oriented segmentation method. The algorithm fully considered multi-spectrum, multi-scale and multi-noises characteristics of high spatial resolu- tion remote sensing image. The details are described as follows. Firstly, an anisotropic diffusion filter was used for image smoothing, because this technology can both remove the noises and maintain edges and other impor-tant details information of the input image. Secondly, in order to take into account the multi-scale characteristics of remote sensing images, multi-scale morphology gradient was extracted because of its good combination of the advantages of large structural element and small structural element, and then H-minima technology was used to extract tags of gradient image for the latter marker-based watershed algorithm. Finally, an improved fast re- gion-merging algorithm was proposed to achieve the multi-scale segmentation. This paper elaborated the pre-pro-cessing filtering, multi-scale gradient, marking extraction and multi-scale region merging aspects, and the experi-ments showed that the proposed segmentation m