高分辨率遥感影像的建筑物轮廓信息提取方法  

Extraction of building contour from high resolution images

作　　者：周绍光[1] 孙金彦[1] 凡莉[1] 向晶[1] 陈超[2] 

机构地区：[1]河海大学地球科学与工程学院测绘科学与工程系,南京210098 [2]江苏省基础地理信息中心,南京210013

出　　处：《国土资源遥感》2015年第3期52-58,共7页Remote Sensing for Land & Resources

基　　金：国家自然科学基金项目"高分辨率遥感影像中城市道路网的提取方法研究"(编号:41271420/D010702)资助

摘　　要：针对高分辨率遥感影像分割方法提取的建筑物边缘不准确和不规则等问题,提出了一种新的边缘轮廓信息提取方法:首先,通过一维Gabor滤波器获取建筑物的角度纹理特征,并结合光谱特征构造待分割的特征矢量,在运用高斯混合模型(Gaussian mixture model,GMM)构造图的基础上,利用图割法(graph cuts)获取建筑物候选点,经数学形态学处理得到建筑物斑块;然后,根据Radon变换检测建筑物主方向,构建最小二乘匹配模板,并利用该模板在建立的轮廓缓冲区内精确地提取建筑物拐角点;最后,连接拐角点,完成了轮廓信息的提取。采用合成图像和高分辨率遥感影像提取建筑物轮廓信息的实验证明了该方法的可行性。Since the building profile obtained by segmentation or other methods has the disadvantages of inaccuracy or irregularity, this paper presents a new approach to extract the outlines of buildings： Firstly, images are preprocessed by combining spectral characteristics and multi -angle texture characteristics obtained by one dimensional Gabor filter of images to form characteristics to be segmented. On the basis of the construction of graph by Gaussian mixture model, the candidate points of the building can be determined by graph cuts, and the building blob can be obtained by mathematical morphology. Then according to segmentation objects, the main direction of the building is detected by the Radon transform, the least square matching templates are created, and the comer points are extracted precisely in the outline buffer zone. Finally, the accurate comer points are connected to constitute the outlines of the building. This method was tested by using synthetic image and high resolution images. The experimental result proves that this method is feasible.

关 键 词：最小二乘模板匹配 高分辨率影像 建筑物轮廓 图割 影像分割 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]