中文摘要：

基于面向对象特征影像分析的思想,提出了一种结合建筑物和阴影对象邻近关系特征的建筑物识别提取方法。在多尺度影像分割的基础上,利用对象的光谱和形状等特征,建立简单的分类决策树,提取粗略的建筑物候选区和相对准确的阴影区。计算相邻近阴影对象和建筑物对象的关系特征,建立简单的知识规则,即可从建筑物候选区中消除广场等噪音,获得准确有效的建筑物目标信息。通过Qu ickB ird卫星影像的实验,证明了该方法在高分辨率卫星影像建筑物目标识别中具有相当的适用性和准确性。

英文摘要：

A practical method is proposed in this paper for building extraction from remote sensing images with high spatial resolution. Relevant features between building and its neighboring shade were used to establish the method. The steps of the approach were as follows： First, the high -resolution merged image was constructed, which combined the Grey Level Concurrence Matrix （GLCM） homogeneity texture feature and the normalized difference vegetation index （NDVI） segmented by arithmetic of multi -resolution segments and two scale feature unit layers. Second, water and land were separated by a threshold of normalized difference water indices （NDWI） based on the larger scale object and then the underlying building region was extracted by the decision rule based on spectral and shape features of the object from the land region according to the larger scale object layer. Third, shade was extracted by the knowledge rule based on the mean value of the near infrared band of objects in the small - scale objects layer. After that, the class -related feature neighboring the shade was defined. Finally, a building was extracted from the building region by searching feature unit objects neighboring the shade. The experimental result based on the QuickBird image shows that the proposed method is very effective and suitable for building extraction from high spatial resolution remote images.