一种新的图像语义自动标注模型

摘要

A focal point of recent CBIR research is automatic image annotation according to the relationship between low-level image features and high-level semantic concepts.An image retrieval framework is introduced,and a novel image annotation method via semantics link network is proposed. By accumulating and learning users' preferences, image semantics are mined. Thus image annotations are obtained by utilizing the proposed algorithm,and subsequently image semantics and annotations are updated automatically in user interactions.Meanwhile,to work out the mistaken annotation propagation problem,a word-relatedness analysis algorithm is also discussed. The annotation approach proposed is tested on the Corel image database,and promising results are achieved.%根据图像低层特征和高级语义间的对应关系,自动进行图像语义标注是目前图像检索系统研究的热点.简要介绍了基于图像语义连接网络的图像检索框架,提出了一种基于该框架的图像自动标注模型.该模型通过积累用户反馈信息,学习并获得图像语义,从而进行自动的图像标注.图像语义及标注可以在与用户交互过程中得到实时更新.还提出了一种词义相关度分析的方法剔除冗余标注词,解决标注误传播的问题.通过在Corel图像集上的对比实验,验证了该方法的有效性.