摘要： 现有基于属性的图像检索主要依赖于预标签属性,使用户只能通过预定义的属性来搜索目标。基于扩展属性的方法则可使用户输入与预标签属性相关的查询词,而非仅选择预定义属性。为此,设计基于网络检索量的扩展属性学习方法。利用Wiktionary挖掘扩展属性,将其与WordNet所得结果相结合,使用由百度指数和谷歌趋势获得的预定义属性及其相应扩展属性的相对平均检索量度量用户偏好,并通过一致性度量方法验证扩展属性的可靠性。实验结果表明,该方法可有效提高图像检索性能。

关键词: 图像检索, 扩展属性, 语义关系度量, 相对平均检索量, 用户偏好, 一致性度量

Abstract: Existing attribute-based image retrieval principally relies on pre-labeled attributes,which restricts users to use only the pre-defined attribute to search the intended targets.Extended attribute-based methods turn users from passively choosing pre-defined attributes to actively inputting query words which are pertinent to the pre-labeled attributes.This paper proposes an extended attribute learning method based on Web search amount.It uses Wiktionary to mine extended attributes and combines the results with that of WordNet.After that,it exploits relative average retrieval amount of attributes obtained from Baidu Index and Google Trends to measure user preference,then adopts a consistency measure method to validate the reliability of the extended attributes.Experimental results demonstrate the significant image retrieval performance improvements of the proposed method.

Key words: image retrieval, extended attribute, Semantic Relation(SR) measure, relative average retrieval amount, user preference, consistency measure

中图分类号: 

• TP391