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  

Rivers on Fuzzy Aerial Images Extracted by Retinex and Improved Mini Spanning Tree Segmentation Algorithm

作　　者：杨楠[1] 王卫星[1,2] 赵祥模[1] 

机构地区：[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

出　　处：《山东农业大学学报（自然科学版）》2017年第6期890-896,共7页Journal of Shandong Agricultural University：Natural Science Edition

基　　金：国家自然科学基金(61170147);陕西省科学技术研究发展计划项目(2013KW03);西安市科技计划项目(CX1252(8));中央高校基本科研业务费专项资金项目(2013G2241019)

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。This paper presents an algorithm combining Retinex with improved minimum spanning tree segmentation, whichcan accurately extract the river from fuzzy aerial image. Firstly the image is enhanced by multiscale Retinex （MSR）algorithm, then the main rivers are detected based on the shape prior and the improved minimum spanning tree segmentationalgorithm. Different fuzzy aerial images are tested, and four well-known image enhancement algorithms and four commonimage segmentation algorithms are compared. The testing results show that without the image enhancement of MSR, thesingle minimum spanning tree algorithm cannot accurately extract the river from the fuzzy aerial image. Compared with thetraditional algorithm, the new algorithm has obvious detection effect on fuzzy or uneven river images.

关 键 词：RETINEX 最小生成树 图像提取 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]