基于多尺度继承性SAR图像分割算法

SAR image segmentation based on multiscale inherited information

针对高分辨率SAR图像的分割问题,提出一种基于多尺度继承性的分割算法。该算法综合利用图像的宏观和微观特征,将传统的单尺度信息处理技术纳入尺度不断变化的动态分析框架中,更容易获得图像的本质特征。同时,使用异性扩散方程获得多尺度图像序列,采用一种由粗尺度到细尺度的分割策略,先进行粗尺度分割,然后以此分割结果来引导较细尺度层的分割。分割过程中采用迭代自组织的数据分析算法自适应地确定每一层分割的区域个数,较好地建立尺度之间的分割继承关系。该分割算法可以满足不同图像处理任务的需求,也更加符合人的认知过程和视觉处理系统。

This paper proposed a method based on inherited multiscale information for SAR image segmentation.It combined image＇s macroscopical and microcosmical features together,and introduced the traditional single scale processing technique into the dynamic changing multiscale analyzed framework,which made it easy to obtain image essential features.Adopted anisotropic diffusion equation to get multiscale images sequences.Used the segmentation strategy from coarser scale to finer scale.Coarser scale was easy to be segmented and the segmentation results could lead finer scale segmentation.Employed iterative self-organizing data analysis techniques algorithm in segmentation process under each scale,and established segmentation inherited relationships of different scales through this process.This method can fulfill the request of different image processing task,and consists with people＇s cognizing process and vision process system.