基于区域生长的林区遥感图像分割方法  

Segmentation Method for Forest Zone Remote Sensing Image Based-RG Algorithm

作　　者：李政国[1] 李春干[1] 代华兵 

机构地区：[1]广西大学林学院,南宁530005 [2]广西林业勘测设计院,南宁530011

出　　处：《广西林业科学》2008年第2期66-70,共5页Guangxi Forestry Science

基　　金：国家自然科学基金(30371159);广西壮族自治区林业局科学基金(200246)共同资助

摘　　要：依据林区遥感图像的特点,利用区域生长法对高分辨率遥感图像进行分割。首先将图像看作多维矩阵结构,建立种子标记矩阵,以标记未处理的像素点作为种子,合并其周围连通像素;其次建立连续自动种子选取机制,遍历整个图像区域,将图像分割为若干区域;最后,采用最小欧氏距离方法合并较小的破碎区域,有效地解决了区域生长对种子点依赖性强,并且容易过度分割的问题。通过实地对比验证评价,分割效果较好。According to the remote sensing image characteristic of forest region, high resolution forest region remote sensing image was segmented by using region growing （RG） algorithm. Firstly, the image was regarded as a multidimensional matrix, a symbol matrix was created for marking those processed pixels and the unmarked pixel become new seed, pixels surrounding it were merged into a region; Secondly, auto-selected seed mechanism was developed to traverse the whole image, and it was divided into many homogeneous regions; Finally, euclidean distance was applied to merge undesired small region, which solved several difficult problems that RG algorithm depended too much on seed pixels and produced many undesired small region. Comparing to forest region reality on the study spot, the segmentation result proved to be good.

关 键 词：区域生长 林区 遥感 图像分割 

分 类 号：S757.2[农业科学—森林经理学]