一种具有相关反馈的图像检索方法  

AN APPROACH TO IMAGE RETRIEVAL WITH RELEVANCE FEEDBACK

作　　者：韦娜[1] 王涛[2] 

机构地区：[1]长安大学信息工程学院,陕西西安710064 [2]武警工程学院基础部,陕西西安710086

出　　处：《计算机应用与软件》2007年第12期168-170,共3页Computer Applications and Software

摘　　要：图像底层特征和高层语义之间存在着巨大的语义鸿沟。受限于图像理解技术的发展水平和对认知的理解水平。目前,对图像语义的描述还无法由计算机自动建立。要克服语义鸿沟,需引入相关反馈机制。特征提取采用结合空间信息的颜色一致直方图方法,并建立了基于方差分析的权值调整方法进行反馈调节,有效地提高了图像检索准确率。The low-level features do not have a direct link to the high-level concepts. However,what current computer vision techniques can automatically extract from images are mostly low-level features. To narrow down the semantic gap, relevance feedback mechanism is needed. A new image feature called color coherent histogram is adopted, which incorporates the spatial information and uses it for image indexing and comparison. A weighted distance approach to relevance feedback based on the analysis of variance is built. It is proved that the approach can improve the image retrieval precision effectively.

关 键 词：基于内容的图像检索 颜色一致直方图 相关反馈 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]