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。
Abstract:
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 corner points are extracted precisely in the outline buffer zone. Finally, the accurate corner 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.
下载: