基于Retinex和改进的最小生成树分割提取模糊航空图像中的河流.PDF

山东农业大学学报 自然科学版( ),2017,48(6):890-896 VOL.48NO.62017 Journal of ShandongAgriculturalUniversity (Natural ScienceEdition ) doi:10.3969/j.issn.1000-2324.2017.06.017 基于Retinex和改进的最小生成树分割提取模糊航空图像中的河流 1* 1,2 1 杨 楠 王卫星 赵祥模, , 1. 长安大学 信息工程学院, 西安 陕西 710064 2. 瑞典皇家工学院, 瑞典 斯德哥尔摩 999027 摘 要: 本文提出了一种Retinex 和改进的最小生成树分割相结合的算法，能够准确地从模糊航拍图像中提取出河流。 首先，通过多尺度Retinex(MSR)算法增强图像，然后基于形状先验和改进的最小生成树分割算法检测出主要河流。 分别测试了各种不同模糊航拍河流图像，比较了四种众所周知的图像增强算法和四种常用的图像分割算法。结果表 明：若没有MSR 的图像增强，单一的最小生成树算法无法准确地提取模糊航拍图像中的河流；新算法对模糊或光照 不均的河流图像，与传统算法比较有着明显的检测效果。 关键词:Retinex; 最小生成树 图像提取; 中图法分类号:TP751 文献标识码:A 文章编号:1000-2324(2017)06-0890-07 Rivers on FuzzyAerial Images Extracted by Retinex and Improved MiniSpanningTreeSegmentationAlgorithm 1* 1,2 1 YANGNan ,WANGWei-xing ,ZHAOXiang-mo 1.Schoolof InformationEngineering/Changan University,Xi’an 710064,China 2.RoyalInstituteof Technology, Stockholm 999027,Sweden Abstract: This paper presents an algorithm combining Retinex with improved minimum spanning tree segmentation, which can accurately extract the river from fuzzy aerial image. Firstly the image is enhanced by multiscale Retinex (MSR) algorithm,then themain rivers are detectedbased on the shapeprior andthe improved minimum spanningtree segmentation algorithm. Different fuzzy aerial images are tested, and four well-known image enhancement algorithms and four common image segmentation algorithms are compared. The testing results show that without the image enhancement of MSR, the single minimum spanning tree algorithm cannot accurately extract the river from the fuzzy aerial image. Compared with the traditionalalgorithm,thenew algorithmhasobviousdetection effect on fuzzy oruneven river images. Keyw