基于聚类的超像素分割算法研究  

Research on Superpixel Segmentation Algorithm Based on Clustering

作　　者：覃晓[1] 覃正优 元昌安[1] 伍永 QIN Xiao;QIN Zheng-you;YUAN Chang-an;WU Yong(College of Computer and Information Engineering,Guangxi Teachers Education University,Nanning 530299,China)

机构地区：[1]广西师范学院计算机与信息工程学院,广西南宁530299

出　　处：《广西师范学院学报（自然科学版）》2018年第1期66-71,共6页Journal of Guangxi Teachers Education University(Natural Science Edition)

基　　金：广西自然科学基金(2016GXNSFAA380209;2014GXNSFDA118037);广西科技计划项目(桂科AB16380272;桂科攻14124005-02-07);南宁市科技开发计划项目(#20175177)

摘　　要：超像素分割是一种重要的图像预处理工具,有关这方面的研究是近年来计算机视觉领域的研究热点。该文对近年来基于聚类方法的超像素分割算法综合分析,并重点就2000年以来三种重要的基于聚类方法的超像素分割算法进行了实验对比,从分割效果和运行效率等方面进行了分析和比较。实验效果显示:DBSCAN超像素生成算法在执行效率和边缘贴合度上都优于其他两种算法,这对于计算机视觉和图像处理领域需要进行快速准确图像分割的应用场景具有重要的参考作用。As an important preprocessing tool of many applications in the field of image processing,superpixel segmentation algorithm has attracted substantial attention in recent years.This paper has summarized and discussed types of superpixel segmentation methods based on the clustering in the past few years,emphatically comparing and analyzing the algorithms of three important superpixel segmentation algorithms in terms of the theory,the segment results and the efficiency,which is based on clustering since 2000.The experimental results show that the DBSCAN method achieves better performance than the other two superpixel segmentation methods in terms of both efficiency and boundary adherence,which gives an important reference for applications that require fast and accurate image segmentation in the field of computer vision and image processing.

关 键 词：超像素 图像分割 聚类 评价指标 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]