遥感图像中建筑物自动识别与标绘方法研究  

Technique for automatic building recognition and mapping in remote sensing images

作　　者：李海月[1] 王宏琦[1] 陆见微[1] 林殷[1] 王富森[2] 

机构地区：[1]中国科学院电子学研究所,北京100080 [2]吉林省吉林市中心医院,吉林132011

出　　处：《电子测量技术》2007年第2期15-20,共6页Electronic Measurement Technology

摘　　要：为了从高分辨率城区遥感图像中自动提取出建筑物目标并对其位置和外形进行标绘,本文提出了一种建筑物自动识别与标绘的方法。该方法首先利用纹理特征对图像进行有效分割,产生出建筑物的候选区域。然后再判断这些区域是否为真正的建筑物区域。最后,对建筑物区域使用一种网格匹配的方法将目标建筑物重建为规则的多边形,以达到建筑物自动标绘的目的。本文用上述方法对IKONOS、QuickBird卫星的高分辨率光学图像进行了实验,实验结果证明了该方法有着较高的识别率、较好的准确性和鲁棒性,具有实用价值。In order to extract and map building targets in high resolution urban remote sensing images automatically, this paper proposes a new solution of automatic building recognition and mapping in remote sensing images. Firstly, this solution uses the texture information to segment the image effectively, thus the candidate areas of buildings can be gotten. Then, the real building areas are ascertained by judging their dominant lines＇ length. Finally, a grid matching method is used to reconstruct the target buildings into regular polygons. This solution has been used on IKONOS and QuickBird images, the experimental results prove that this solution has a good precision and robustness.

关 键 词：目标识别 自动标绘 图像纹理分割 灰度共生矩阵 HOUGH变换 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]