基于活动轮廓模型的高分辨率影像河道边界提取算法研究

河( 航) 道边界定位是河( 航) 道测量的重要内容, 对于内河航运交通、水利工程建设以及水资源开发利用等有着非常重要的意义。在目前遥感影像水体信息提取的研究基础上, 结合计算机视觉领域中的可变形模板轮廓匹配技术的优点, 同时考虑高分辨率影像上河道边界的光谱特征, 提出了一种基于活动轮廓模型的高分辨率影像河道边界提取算法。数据源选用四川宜宾地区的QuickBird影像, 采用本算法对初始轮廓进行变形, 匹配河道边界的真实位置。对于位于真实边界附近的初始轮廓, 拟合的边界线上落在距离真实边界两个像素内的点数达到70% , 5 个像素以内的点数达到90%以上。结果说明引入计算机视觉领域中比较成熟的技术将给高分辨率遥感影像的信息提取开辟新的途径。

A River Bank Extracting Algorithm Based on Active Contour Model from High-resolution Satellite Images

Positioning river bank is very important for channel surveying and gauging,which has significant meaning in freshwater navigation,hydraulic and hydroelectric engineering building,water resource exploitation and so on.On the basis of current research on river extraction from satellite imagery and the application of deformable template model in computer vision,the paper puts forward an active contour model to extract river bank from high resolution satellite imagery.The model can take both spectrum and shape character into account and is tested on QuickBird images with 0.6m spatial resolution in Yibin City,Sichuan province.Rivers extracting from Landsat TM imagery and small scale maps or manual river sketch are input as the initial estimate of river bank,and these contours will be shaped as the river bank in the QuickBird image using the model.About 70 percent of the total points in the final contour locate within two-pixel distance from the accurate position,and above 90 percent are situated within five-pixel distance when the initial contour is close to the river.The result points out that introducing the mature technology in compute vision research field is very promising for feature extraction from high-resolution satellite imagery.