基于彩色形态筛的建筑物主体轮廓提取算法

摘要

A new automatic extraction method for main building contour is proposed. It applies convex color sieves (CCS) of small-scale to construct a city-building Bayesian model. From the maximum a posterior (MAP) of the angles, the positions of three vanishing points can be determined. Then, the main building contour is extracted according to edge-pixels cluster resuits. The algorithm is applied to single image without involving processing stage such as edge detection or Hough transform. Experimental results show that estimating the constructed grid can automatically obtain the vanishing points and extract the main parallelepiped border of building.%针对常见的城市建筑物图像,提出了一种由单幅图像自动完成建筑物主体轮廓提取的算法.算法利用基于凸包彩色形态筛的多尺度性,设计了城市建筑物场景贝叶斯概率统计模型,由最大化后验概率(Maximum a posterior,MAP)估计消隐点在图像中的投影,利用边缘像素分类的结果获得建筑物的主体轮廓.该算法可用于单幅图像,不需要边缘检测和Hough变换等处理.实验结果证明,通过估计该场景结构可以自动获得消隐点在图像中的投影,进而提取建筑物的平行六面体主体轮廓.