一种基于概率粗糙集模型的图像语义检索方法  

An image semantics retrieval method based on probability rough set model

作　　者：徐久成 李晓艳[1] 孙林[1] 

机构地区：[1]河南师范大学计算机与信息技术学院,河南新乡453007

出　　处：《南京大学学报（自然科学版）》2011年第4期438-445,共8页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(60873104;61040037);河南省省教育厅自然科学基金(2008B520019)

摘　　要：针对目前图像数据信息量大、检索不易和人们对图像检索习惯于对图像概念语义进行检索的难题,本文将概率粗糙集理论和图像的语义标注技术引入图像的信息检索中,提出了一种基于朴素贝叶斯理论和概率粗糙集模型的图像语义信息检索模型.首先,针对图像库中的图像构造精确标注词空间,并通过朴素贝叶斯理论对图像进行精确标注和模糊加权标注.将概率粗糙集模型和朴素贝叶斯理论的后验概率相结合,计算每对图像标注词的条件概率和模糊条件概率,并求得每个标注词的支持集和被支持集,在此基础上,计算每个标注词的支持集和被支持集的上、下近似,并通过上、下近似构造图像的语义相似度计算方法,之后计算待查询图像的查询特征与图像库中图像之间的语义相似度,并根据相似度的大小给出检索的排序和输出.最后,给出一个简单的仿真实验,实验结果表明该方法是有效可行的.In view of the problem that the present image data contains too much information,not easy to retrieve and people is used to searching the image semantic concept in the process of image retrieval,this paper introduces the probabilistic rough set model and image semantic annotation technology into the image information retrieval and puts out a kind of image information retrieval model based on Nave Bayesian and probability rough set model.Firstly,the images are crisp annotated and weighted fuzzy annotated by Nave Bayesian after building crisp annotation space for the image library information.Secondly,the conditional probability and fuzzy conditional probability are computed for each pair of annotations by combining the posterior probability of Nave Bayesian with the probabilistic rough set model,after that the support set and supported set of each annotation are computed,the upper and lower approximation of each annotation are calculated for the support set and supported set,and the image semantic similarity calculation method is built by the upper and lower approximation,then the semantic similarity between query characteristics and the characteristics of image library is calculated,and the retrieval results are output by ordering according to similarity.Finally,taking the images in Corel image library for an example,the paper give a simple simulation experiment which shows the algorithm is feasible and effective in practice.

关 键 词：概率粗糙集 图像语义 自动标注 朴素贝叶斯 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]