基于正负例反馈的最大似然估计图像检索算法  

Maximum Likelihood Estimation Image Retrieval Algorithm Based on Positive and Negative Examples Relevance Feedback

作　　者：何丽[1] 杨文君[2] 

机构地区：[1]北方工业大学计算机系,北京100144 [2]牡丹江师范学院计算机科学与技术系,牡丹江157012

出　　处：《武汉理工大学学报》2009年第18期142-146,共5页Journal of Wuhan University of Technology

基　　金：北京市教委科技发展计划项目(KM200810009009);校科研青年教师发展基金(0815)

摘　　要：为了提高图像检索的性能,提出了一种基于相关反馈的最大似然估计图像检索算法。该算法根据人眼的视觉感知原理,对低频与高频成分区分处理,以频率层色彩和形状为图像特征,根据用户少量的正负实例反馈信息,采用最大似然估计自适应地调整图像特征权值。实验结果表明,提出的算法与基于支持向量机的相关反馈算法SVMRF相比,查准率至少提高10%,查全率至少提高8%。To improve the performance of image retrieval, a maximum likelihood estimation image retrieval algorithm is proposed based on relevance feedback. According to visual perception theory, the proposed algorithm distinguishes high and low frequency component and adopts color and shape of frequency layer as image features, and adaptively adjusts the weights of features by maximum likelihood estimation with the feedback information of few positive and negative examples. Detailed experimental results show that, compared with the feedback algorithm SVMRF based on support vector machine （SVM）, the proposed algorithm improves accuracy ratio no less than 10%, recall ratio no less than 8%.

关 键 词：图像检索 正负例 相关反馈 最大似然估计 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]