图像检索中基于向量转移的相关反馈算法的研究与改进  

Research and improvement of relevance feedback based on vector transfer in image retrieval

作　　者：郑秋梅[1] 蒋晓红[1] 杨发科[1] 

机构地区：[1]中国石油大学计算机与通信工程学院,山东东营257061

出　　处：《计算机工程与设计》2008年第17期4501-4503,4517,共4页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2007AA09Z301)

摘　　要：相关反馈方法是对基于内容图像检索系统的有效改进。对图像检索系统中基于向量转移的相关反馈算法进行了研究。针对传统向量转移算法速度较慢、误差较大的不足,提出了一种基于非查询向量转移相关反馈算法的改进办法,通过提高向量转移的速度和精度,有效的利用用户的反馈信息,从而加快相关向量的收敛速度。经过公用图像数据库上的实验测试表明,该方法能有效减少反馈次数,提高检索效率。Relevance feedback is an efficient improvement to the content-based image retrieval system. Relevance feed-back based on vector transfer in image retrieval is studied. Aimed at shortage of low speed and quite big error of vector transfer arithmetic, a new method of relevance feedback based on non-query vector transfer is presented. Increasing the speed and precision of vector-transfer can make use of feedback information efficiently and quicken convergence of relevant vector. The results of the experiments on the public image-database show that the method could reduce the time of feedback and improve the efficiency of retrieval effectively.

关 键 词：图像检索 相关反馈 向量转移 速度 精度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]