基于随机投影深度函数的非典型道路中心线提取

Non-typical road center line extraction based on random projection depth function

提出一种结合随机投影深度函数和路径形态学的非典型道路中心线提取方法.随机投影深度函数可有效区分RGB彩色空间中数据集的数据中心与离群值,从而充分利用非典型道路光谱特征的异质性及其与图像背景光谱特征的差异性来凸显具有离群值光谱特征的道路.利用随机投影深度函数对彩色遥感图像进行由中心向外的排序,生成深度场图像,并采用直方图阈值分割方法二值化深度场图像;利用路径形态学粗提取道路区域,去除独立斑块、连接断路来精细化道路区域;利用形态学细化操作提取道路中心线,并去除细枝以得到精准的道路中心线.实验采用Ikonos和ZY-3图像,结果表明:该方法可有效提取非典型道路中心线,其提取道路中心线的完整率、正确率和检测质量平均达到94.31%,94.16%和90.02%.

Random projection depth function and path opening and closing were introduced to extract the center lines of non-typical roads.In RGBcolor space,the random projection depth function can effectively distinguish the center and outlier of a data set,so that it can highlight the roads with outlier spectral characteristics by making full use of the heterogeneity of nontypical road spectral characteristics and its difference from image background spectral features.Firstly,random projection depth function was adopted to provide an ordering for the spectral characteristics of a color image in RGBcolor space,which is from the center to the outward and the median is the deepest point of the data set.Depth field image was generated according to the ordering results,and histogram threshold segmentation was used in the image binary progress.Secondly,path opening and closing was applied to extract the rough road network area,while independent plaque and connection of open circuit were removed to refine the road area.Finally,morphological thinning algorithm was applied to extract the road center lines and a refined operation was performed to get the accurate road center lines.High resolution remote sensing images from Ikonos and ZY-3 sattellites were tested.The experimental results show that the proposed method can effectively extract the non-typical road center lines.The average integrity rate,accuracy,proof mass can reach 94.31%,94.16% and 90.02%.