基于形状先验的建筑物几何参数提取方法

摘要

Due to algorithms for buildings' geometric parameters extraction from single high remote sensing image usually suffer from disturbances of background, noise and intensity similarity, resulting in wrong extraction results. In this paper, a novel variational level set model is proposed based on shape prior which integrates edge, gray, and shape prior including roof and facade, to extract buildings' geometric parameters from a single high resolution remote sensing image. Experimental results show that the proposed method can extract buildings' geometric parameters accurately. Moreover, it has strong robustness with respect to the disturbance of facade due to the more sufficient utilization of the whole shape prior.%基于单幅高分辨率遥感图像的建筑物几何参数提取结果的准确性,容易受图像背景、图像噪声以及灰度分布相似性的干扰,形成错误的提取结果.针对这一问题,该文提出一种新的基于形状先验的变分水平集提取方法,该方法同时使用图像边缘信息、区域灰度信息以及包含屋顶和侧立面的形状先验信息,实现单幅遥感图像中建筑物的几何参数的提取.实验结果表明,该方法能够更加准确地提取建筑物,最后得到的几何参数比较接近真值,并且由于更加充分地使用了全局形状信息该方法能更好地抵制侧立面的干扰,具有很强的鲁棒性.