基于NMI特征的遥感影像线性迭代聚类超像素分割算法  

SLIC super-pixel segmentation algorithm base on NMI features used in remote sensing image

作　　者：李静 LI Jing(Beijing Institute of Tracking and Telecommunications Technology,Beijing 100094,China)

机构地区：[1]北京跟踪与通信技术研究所,北京100094

出　　处：《光学精密工程》2022年第6期734-742,共9页Optics and Precision Engineering

基　　金：粤港关键领域重点突破项目(No.2004A10403021);广东省攻关项目(No.2006A10401006)。

摘　　要：针对现有基于简单线性迭代聚类(SLIC)的超像素分割算法用于细节丰富的遥感图像处理时,存在的易受噪声干扰、过分割问题,本文提出一种结合超像素块之间基于归一化转动惯量(NMI)特征的相似性度量的遥感影像分割方法,对分割效果进行改善。本文首先利用引导滤波算法对影像进行平滑处理,去除椒盐噪点;再通过现有的线性迭代聚类算法对影像进行像素级分割,生成初始的超像素;进而确定出微小超像素块,然后计算其与相邻超像素块的相似性度量值,将其合并入差异性最小的相邻超像素块,达到分割影像的目的。本文方法在传统分割算法基础上降低了超像素对噪声的敏感性,提高了影像分割的精度。实验表明,论文提出算法可将测试遥感图像的分割超像素块数量由4 171减小为282,微小超像素块数量减少60%以上,有效降低噪声点的影响,改善以往算法存在的过分割缺陷。The state-of-the-art super-pixel segmentation algorithm based on simple linear iterative clustering(SLIC)has the problem of over segmentation and discontinuity when processing remote sensing images with extensive details. Here,we propose a remote sensing image segmentation method that combines the NMI-based similarity measure between super-pixel blocks to improve the segmentation effect. First,a guided filtering is used to smooth the pepper noise in the image. Second,the image is segmented at a pixel level using the SLIC algorithm to generate initial super-pixels. Third,to achieve image segmentation,the micro super-pixels are determined based on some criterion and then merged into the adjacent super-pixel blocks with the least difference by calculating the similarity measure with its adjacent super-pixel blocks.This paper’s method reduces the sensitivity of super-pixel to noise and improves the precision of image segmentation compared with traditional segmentation algorithms. The experimental results indicate that the proposed algorithm reduced the number of segmented super-pixel blocks from 4 171 to 282 and reduced the number of micro super-pixel blocks by more than 60%. It also reduced the influence of noise points and improved the over segmentation defects of existing algorithms.

关 键 词：简单线性迭代聚类 超像素 区域合并 归一化转动惯量 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]