基于主色选择的CBIR检索  

CBIR RETRIEVAL BASED ON DOMINANT COLOR SELECTION

作　　者：袁昕[1] 吴春明[1] 朱淼良[1] 王东辉[1] 

机构地区：[1]浙江大学人工智能研究所,杭州310027

出　　处：《计算机研究与发展》2002年第9期1120-1126,共7页Journal of Computer Research and Development

基　　金：国家自然科学基金 ( 6 9875 0 15 );浙江省自然科学基金 ( 11110 1N2 980 9)资助

摘　　要：基于内容的图像检索 (CBIR)是多媒体检索研究的前沿课题 .利用颜色特征作为索引进行图像检索是最重要的技术 .在提取图像主要颜色特征的基础上 ,进一步提取了相应的主色空间分布信息——主色矩特征 ,作为图像库的索引 .在改进加权二次型相似性度量方法的基础上 ,提出了相应的主色多特征相似性度量方法 .由于用户对图像中不同的主色具有不同的检索要求 ,提出了主色调选择的用户模型 ,用于更精确的图像检索 .实现了 WWW发布方式的 CBIR原型系统 ,实验结果表明加入主色选择使得图像检索的效果更好 .Content-based image retrieval (CBIR) is an advanced project in the recent research area of multimedia retrieval. Indexing by color features is the important technology of image retrieval. In this paper, a new kind of image feature, dominant color moment (DCM), which reflects spatial information of images, is taken into account on the basis of the image feature of dominant color (DC). A method of similarity measurement is discussed based on the improvement of quadric form distance function. A kind of interactive model between man and system, called dominant color selection, is also proposed for precise retrieval according to different requirements imposed by different dominant colors in images submitted by users. Such a system shows good effect in the CBIR experimental system implemented by means of WWW distribution.

关 键 词：主色选择 CBIR检索 主色矩 图像检索 图像数据库 计算机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]