基于RSOM和KNN的图像内容检索  

Content Based Image Retrieval Using RSOM and KNN

作　　者：谭立球[1] 谷士文[1] 夏胜平[2] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410075 [2]国防科学技术大学ATR重点实验室,湖南长沙410073

出　　处：《微电子学与计算机》2007年第10期58-61,64,共5页Microelectronics & Computer

摘　　要：RSOM树是一种以SOM网络为节点的层次化聚类树,可以实现将空间上彼此靠近的数据点聚合在树中相同的节点中,它可作为图像索引支持最近邻检索,高效且不必线性扫描高维、海量的数据库。通过提取图像低层特征构成每个图像的特征向量,在此基础上构造基于RSOM树的图像特征索引库,并运用KNN检索实现基于样例的检索。实验证明,无论从特征空间维数、检索性能、样本容量方面都具有良好的性能。RSOM tree, composed of SOM nets and a special kind of cluster tree, is a hierarchy of clusters and sub clusters which incorporates the cluster representation into the index structure, it provides a practical solution to index clustered image data sets and supports the retrieval of the nearest-neighbors effectively and efficiently without having to linearly search a high-dimensional large database. Based on the image low level feature including color, texture and contour, RSOM image tree is constructed and an retrieval example using KNN is present. The excellent performance of the method has been tested in dimension, performance and capacity.

关 键 词：图像内容检索 高维数据 KNN RSOM树 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]