摘要
解决语义鸿沟必须建立图像低层特征到高层语义的映射,针对此问题,本文提出了一种基于词汇树层次语义模型的图像检索方法.首先提取图像包含颜色信息的SIFT特征来构造图像库的特征词汇树,生成描述图像视觉信息的视觉词汇.并在此基础上利用Bayesian决策理论实现视觉词汇到语义主题信息的映射,进而构造了一个层次语义模型,并在此模型基础上完成了基于内容的语义图像检索算法.通过检索过程中用户的相关反馈,不仅可以加入正反馈图像扩展图像查询库,同时能够修正高层语义映射.实验结果表明,基于该模型的图像检索算法性能稳定,并且随着反馈次数的增加,检索效果明显提升.
Creating a semantic mapping method is important in solving semantic map.This paper proposes a vocabulary tree hierarchical semantic model image retrieval method.First extracting the characteristics of the image,such as the color information,SIFT feature,and then generating the visual vocabulary to describe the image of visual information.Using Bayesian theory map the visual vocabulary to the semantic subject information,then constructing a hierarchical semantic model.So based on this model we realize a semantic content-based image retrieval algorithms.Through the user's relevance feedback,not only can be added to the positive feedback image,but also can be able to amend the high-level semantic mapping.Results perform stability,and with the increase in the number of feedback retrieval results can improve significantly.
出处
《微电子学与计算机》
CSCD
北大核心
2012年第11期172-176,共5页
Microelectronics & Computer
基金
国家自然科学基金(60970015
61003054
61170020)
江苏省高校自然科学研究项目(10KJB520018)
苏州市科技支撑计划项目(SG201257)
关键词
词汇树
语义主题信息
层次语义模型
语义映射
图像检索
vocabulary tree
semantic subject information
hierarchical semantic model
semantic mapping
image retrieval