基于感兴趣区域的图像检索方法  

An Image Retrieval Method Based on Regions of Interest

作　　者：斯白露[1] 高文[1] 卢汉清[2] 曾炜[3] 段立娟[1] 

机构地区：[1]中国科学院计算技术研究所,北京100080 [2]中国科学院自动化研究所,北京100080 [3]哈尔滨工业大学计算机科学与工程系,哈尔滨150001

出　　处：《高技术通讯》2003年第5期13-18,共6页Chinese High Technology Letters

基　　金：863计划(2001AA140221)资助项目。

摘　　要：为了弥补全局特征在描述图像内容上的不足,近年来许多研究者提出了基于区域的图像表示和检索方法。本文提出了一种基于感兴趣区域的图像检索方法,根据认知心理学中选取注视焦点的模型来提取感兴趣区域,利用感兴趣区域的视觉特征和位置关系进行图像检索。该方法既利用了区域的局部特征,又考虑了区域在重要程度上的差别。在60000幅通用图像库上的实验结果表明,这种方法的检索效率要好于现有的基于区域的方法,而且对查询图像的噪声等扰动具有鲁棒性。In order to overcome the deficiencies of global features, recent years researchers have proposed many region based image retrieval methods to include spatial information of an image. This paper describes an image retrieval method based on regions of interest. Using this method, an image is represented by a set of regions of interest, which are extracted through visual attention model. An important aspect of this method is that it not only takes local features into consideration, but also the importance of regions. Experiments on a database of about 60 000 general-purpose images demonstrate that this method provides a remarkable improvement compared with common region based methods. Moreover, this method is robust to image alterations.

关 键 词：感兴趣区域 图像检索 视觉特征 位置关系 图像库 检索效率 图像处理 数据库 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]