计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (20): 155-158.DOI: 10.3778/j.issn.1002-8331.2008.20.047

• 数据库、信号与信息处理 • 上一篇    下一篇

一种新的基于重排序的相关反馈图像检索方法

杨德三,李 明,刘 玲   

  1. 兰州理工大学 计算机与通信学院,兰州 730050
  • 收稿日期:2007-09-27 修回日期:2007-12-07 出版日期:2008-07-11 发布日期:2008-07-11
  • 通讯作者: 杨德三

New relevance feedback based on re-ranking mechanism in content-based image retrieval

YANG De-san,LI Ming,LIU Ling   

  1. School of Computer and Communication,Lanzhou University of Technology,Lanzhou 730050,China
  • Received:2007-09-27 Revised:2007-12-07 Online:2008-07-11 Published:2008-07-11
  • Contact: YANG De-san

摘要: 提出了一种新的相关反馈方法,该方法引入了Rnorm重排序机制。通过计算用户反馈的按个人兴趣排列的期望输出顺序与系统输出图像顺序之间的Rnorm值,来调整各个特征的权重,从而指导下一轮的检索。新方法不需标注,减轻了用户的负担,从而避免了用户是否愿意配合的问题,而且实验表明较Rui方法在性能上有很大提高。

关键词: 颜色, 纹理, 形状, 特征融合, 重排序机制, 相关反馈

Abstract: A new relevance feedback method is proposed in the paper,which based on re-ranking mechanism Rnorm.Rnorm is used to calculate the similarity between system output order and user own-wish order.Then features’ weights are automatically adjusted to guide next retrieval.The new relevance feedback method doesn’t need user’s heavy and complicated tagging work,and experimental results show that it outperforms Rui method.

Key words: color, texture, shape, feature fusion, re-ranking mechanism, relevance feedback