In high resolution remote sensing images, the traditional pixel-based methods are inefficient when extracting information from images. This paper introduced an object-based image analysis method to extract building information. At the beginning, it used a multi-resolution segmentation method to divide the image into segments, which were image objects that implemented further analysis and classification. Through the intra-segment and inter-segment heterogeneity measures, it used an unsupervised optimal scale method to segment the image. In order to get a better segment results, it segmented the under-seg-ment and over-segment regions caused by single scale segmentation with a smaller scale and merged respectively. To classify the image objects, it used the digital surface model ( DSM) derived from LiDAR data and spectral information together to analyze and get properties of buildings, and extracted coarsely by height distribution and the green ratio of objects the building. To get a refined building extraction, it used spatial information such as size and position. In the final result, experment extracted 18 buildings objects from the high resolution urban image. The result shows the method is efficient and feasible.%针对基于像素分析方法不适用于高分辨率影像信息提取的问题,提出一种基于对象的图像分析方法来进行城市建筑信息提取.采用多分辨率图像分割方法得到图像对象,提出非监督的最优尺度判定方法解决单尺度分割造成的欠分割和过分割问题.在对象分类提取过程中,结合LiDAR数据的地形表面高程信息和光谱信息对建筑物进行提取,并利用尺寸、空间位置等信息进行误分类修正.实验区域共提取出18个建筑目标,结果表明所提出的方法有效可行.
展开▼
