基于内容图像检索中相关反馈技术的回顾  

A Survey of Relevance Feedback Techniques in Content-Based Image Retrieval

作　　者：吴洪[1] 卢汉清[1] 马颂德[1] 

机构地区：[1]中国科学院自动化研究所模式识别国家重点实验室,北京100080

出　　处：《计算机学报》2005年第12期1969-1979,共11页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(G1998030502);中国科学院知识创新工程项目资助.

摘　　要：由于相关反馈技术能有效地提高基于内容图像检索的性能,使它成为图像检索系统中不可少的一部分.近年来相关反馈技术的研究正吸引着越来越多的关注,涌现出了许多算法.在简要介绍了基于内容图像检索后,文中讨论了相关反馈的交互过程和其中的重要环节,进一步分析了相关反馈中的学习问题及其特点,根据相关反馈算法所采用的检索模型把算法分为基于距离度量的方法、基于概率框架的方法和基于机器学习的方法,并在这个分类下对近年来有代表性的一些算法进行了分析和探讨,最后展望了相关反馈技术未来的发展方向.Relevance feedback, as an effective approach to boost image retrieval, has become a necessary part of content-based image retrieval system, and attracted much research attention in the past few years. This paper provides a comprehensive survey of relevance feedback techniques described in the literature. After a brief introduction of content-based image retrieval, the interactive process of relevance feedback and its import aspects are discussed. Relevance feedback is further formulized as a supervised learning problem, and its characters are analyzed. Based on the retrieval model adopted in the algorithm, relevance feedback algorithms are categorized into three classes, distance-based approach, probabilistic approach, and machine learning based approach, and various representative algorithms are introduced following this categorization. At last, some promising research directions are also suggested.

关 键 词：相关反馈 基于内容图像检索 监督学习 小样本 用户相关判断 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]