Traditional object extraction methods encounter new challenges as the spatial resolution of remote sensing image increases. For the purpose of building extraction in that circumstance, a new method combining both attribute filtering and context analysis is proposed in this paper. Firstly, attribute filters are performed on the object level, which is defined by the connected pixels with similar attribute. Secondly, according to the prior knowledge of buildings and shadows, preliminary regions are extracted as the building candidates. Thirdly, the context between building and shadow is considered, which leads to the aspect angle, relative position and central distance criteria for the confirmation of buildings from the building candidates. Results on both of the residential and industrial regions demonstrate that the proposed method is effective for building extraction and performs better than the traditional methods.%遥感图像分辨率的提高在带来丰富细节的同时,也给传统的目标提取方法提出了新的挑战.该文根据建筑物目标的属性和上下文关系,提出了一种新的高分辨率遥感图像中建筑物目标提取方法.该方法首先以具有相似属性的连通像素为处理单元,依据地物对象间的属性差异,对遥感图像进行不同尺度的属性滤波；接着根据建筑物和阴影的自身属性特点,对上述属性滤波结果进行粗筛,形成备选目标集；最后,进一步考虑建筑物与其阴影间的上下文关系,从二者的方位角、相对位置和区域距离3方面对备选目标集进行筛选,完成建筑物提取.对居民区和工业区遥感图像的实验结果表明,该方法优于传统方法,可用于提取多种类型的建筑物.
展开▼
