摘要
提出了一种基于快速水平集变换(FLST)的多尺度面状地物提取方法。该方法利用FLST变换将图像分解为灰度水平集,面状体物由于在空间分布和灰度值上的相近,其相关信息很容易整体转移到水平集内;通过一种水平线Min/Max流方法对包含地物信息的形状进行多尺度边缘平滑处理,滤除图像中的细节信息,保持面状地物轮廓基本不变;最后进行重构图像,重构结果具有灰度分布分片恒定的特点,很容易从图像中获取目标地物的分布情况。对遥感影像中的典型面状地物,如农田、水域、积雪等,使用该方法进行了地物提取实验,提取结果与人工方法提取相比,其精确度均达到了90%以上。
A method of multi- scale area features extracting based on Fast Level Set Transformation( FLST) is proposed. An image is decomposed into gray level sets by FLST and the area features are transferred to level sets integrally for the sake of adjacency on spatial distribution and gray scales. Edges of shapes containing feature information are smoothed by a level- line Min / Max flowon different scales for removing details and preserving area features,then the reconstructed image is piecewise constant on gray scales. It is convenient to fetch information of area features from resultant image. Relative experiments are carried out on remote sensing images for farmlands,water areas and snowcovers,whose result compared with that of artificial extraction shows that the precisions are all greater than 90 percentages.
出处
《计算机技术与发展》
2014年第12期167-171,共5页
Computer Technology and Development
基金
国家自然科学基金资助项目(41101425)
江苏省自然科学基金(BK20130864)
南京邮电大学引进人才科研启动基金(NY213056)
南京邮电大学实验室工作研究课题(17032SG1315)