一种利用综合评分的相关反馈图像检索方法  

Relevant Feedback Method for Image Retrieval Using Composite Score

作　　者：陈士胃[1] 黄元元[1] 刘宁钟[1] 

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210016

出　　处：《小型微型计算机系统》2011年第12期2481-2486,共6页Journal of Chinese Computer Systems

基　　金：青年科创基金项目(1004-56v1098)资助;国家自然科学基金项目(60903104)资助

摘　　要：提出一种利用综合评分的相关反馈检索方法.该方法将用户对检索的满意度信息连同反馈记录一起存于反馈日志数据库中,在进行新的检索时,将当前反馈样本集合与通过分析历史反馈信息得到的扩充样本集合一起作为软标签SVM的输入,训练出分类器,并用三种评分的加权和作为图像库图像最终的评分值.对1000幅图像数据库的实验表明,与现有相关反馈方法相比,该文方法在提高反馈效率,充分利用历史反馈信息,克服反馈噪声方面都有明显改善.A relevant feedback method for image retrieval using composite Score is proposed in the paper.First,the information,the retrieval satisfaction and feedback records from users,is stored into the feedback log database.Second,when a new retrieval is launched,the current feedback samples are inputted into the Soft Label SVM to get a classifier,together with expanded samples by analyzing historical feedback information to obtain.Finally,the weighted sum of three scores is calculated as the final score.The experiment result show that compared with existing relevant feedback method,the proposed method has significant improvement in feedback efficiency,the full use of the historical feedback,and feedback noise suppression.

关 键 词：基于内容的图像检索 长期学习 历史反馈信息 相关反馈 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]