计算机应用 ›› 2012, Vol. 32 ›› Issue (04): 1097-1100.DOI: 10.3724/SP.J.1087.2012.01097

• 图形图像技术 • 上一篇    下一篇

图像搜索结果的重叠层次聚类与代表点展现

谷瑞军1,陈圣磊1,陈耿1,2,汪加才1   

  1. 1. 南京审计学院 信息科学学院,南京 210029
    2. 江苏大学 计算机科学与通信工程学院,江苏 镇江 212013
  • 收稿日期:2011-10-08 修回日期:2011-11-13 发布日期:2012-04-20 出版日期:2012-04-01
  • 通讯作者: 谷瑞军
  • 作者简介:谷瑞军(1979-),男,山东菏泽人,讲师,博士,CCF会员,主要研究方向:图像检索、数据挖掘;
    陈圣磊(1977-),男,山东兖州人,讲师,博士,主要研究方向:机器学习;
    陈耿(1965-),男,江苏南京人,教授,博士生导师,CCF高级会员,主要研究方向:计算机审计、数据挖掘;
    汪加才(1962-),男,江苏连云港人,教授,博士,主要研究方向:商务智能。
  • 基金资助:
    国家自然科学基金资助项目;国家社会科学基金

Hierarchical overlapping clustering and exemplar visualization of images returned by search engine

GU Rui-jun1,CHEN Sheng-lei1,CHEN Geng1,2,WANG Jia-cai1   

  1. 1. School of Information Science, Nanjing Audit University, Nanjing Jiangsu 210029, China
    2. School of Computer Science and Telecommunication Engineering, Jiangsu University, Zhenjiang Jiangsu 212013, China
  • Received:2011-10-08 Revised:2011-11-13 Online:2012-04-20 Published:2012-04-01
  • Contact: GU Rui-jun

摘要: 针对图像聚类中面临的高维、准确度低、部分重叠等问题,提出了一种高效的基于链接层次聚类的多标记图像聚类。该方法通过图像距离计算相似度,通过链接聚类检测重叠簇。从而每个图像可能归属于多个簇,使得簇标签的意义更明确。为了检验方法的有效性,对通过搜索引擎检索特定关键词返回的图片数据集进行聚类。结果表明,该方法能有效发现具有重叠划分的簇,且簇的意义比较明确。

关键词: 图像聚类, 链接聚类, 多簇划分, 图像距离

Abstract: To resolve the problems of high dimensionality, low accuracy and overlapping in image clustering, an effective link-clustering based image multiple-cluster partition method was proposed in this paper. This method utilized image distance to measure similarity and identified overlapping clusters by using link-clustering. As a result, an image may be partitioned into multiple clusters, and this multiple-cluster partition makes each cluster more specific compared with others. To validate this method, experiments were carried out on the datasets returned by search engine when searching for some key words. The result shows that the proposed method can find explicit clusters with partial overlapping.

Key words: image clustering, link clustering, multiple-cluster partition, image distance