基于图论的无监督区域遥感图像检索算法研究  

Research on Unsupervised Regional Remote Sensing Image Retrieval Algorithm Based on Graph Theory

作　　者：李丽萍[1] 赵传荣[2] 孔德仁[1] 王芳[1] 

机构地区：[1]南京理工大学机械工程学院,南京210094 [2]安徽工业大学电气与信息工程学院,马鞍山243032

出　　处：《计算机科学》2017年第7期315-317,323,共4页Computer Science

基　　金：国家自然科学基金(11372143);国防科工局基础计量项目(J092013B003)资助

摘　　要：为了改善基于内容的遥感图像检索技术,以遥感图像区域检索为框架,提出了一种新的基于图论的无监督学习遥感图像检索算法。首先,提出的方法用图表为每一幅图像建模,同时结合局部信息和相关的空间结构,提供基于区域的图像代表。将每一幅图像初步划分为不同的区域,再通过属性关系图建模,节点和边界分别代表区域特征和它们之间存在的空间关系。然后,通过评估基于图像的相似点实现最相似于查询图像的图像检索。为匹配相应的图像以及按照图像相似点实现图像检索,采用了结合子图同构算法和光谱图嵌入技术的新型非精确图像匹配策略。实验结果显示,与其他两种无监督遥感图像检索方法相比,所提方法的检索性能明显改善。In order to improve the content based remote sensing image retrieval technology,a new image retrieval algorithm based on graph theory was proposed.First,the proposed method models each image by agraph and combines local information and related spatial structures,which provides the region based image representation.Each image is initially divided into different regions.The nodes and boundaries of the attribute relation graph respectively represent the regional features and the spatial relations between them.Then,image retrieval is achieved based on image similarity.To match the corresponding image and realize image retrieval according to image similarities,a new type of non-accurate image matching strategy is used based on sub-graph isomorphism algorithm and spectral graph embedding technology.The experimental results show that compared with the most advanced unsupervised remote sensing image retrieval methods,the retrieval performance of the proposed method is significantly improved.

关 键 词：图像检索 图论 无监督学习 属性关系图(ARG) 子图同构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]