一种遥感图像建筑物检测新方法

遥感图像中的建筑物检测对于土地规划和地图绘制等具有重要意义。文章针对高分辨率全色遥感图像植被覆盖中隐蔽建筑物检测问题，提出了一种结合屋顶结构信息和纹理信息的快速房屋检测方法。首先利用屋顶边缘的几何关系寻找具有矩形屋顶的建筑物；然后使用形态学方法提取屋顶较亮的建筑物；最后利用局部二值模式（Local Binary Pattern，LBP）描述屋顶的纹理特征，并利用支持向量机（Support Vector Machine，SVM）排除误检区域。通过在全色遥感图像集上的试验证明，所提方法对高分辨率遥感图像植被覆盖区域中的建筑物目标具有较高检测率和较低误检率。

A New Approach to Building Detection in Remote Sensing Images

Building detection from very high resolution satellite images is an important task for land planners and mapping. However, manually locating buildings is a difficult and time consuming process. In this study, a novel automatic building detection approach based on line segment detector （LSD） and local binary pattern （LBP） features under the vegetation environment from very high resolution satellite image is presented. The rectangular buildings can be identified by analyzing the specific geometric relationships between edges of the roof-top. Furthermore, the threshold-based segmentation and morphological operations are used to find light-colored roof-top areas. We compose the results of the above procedures to form the potential areas of roof-tops. Finally, a support vector machine （SVM）, with the help of LBP texture features, is used to remove non-roof regions. The experiment results demonstrate the effectiveness of the method.