两种近似EMD的图像检索方法  

Two methods approximating EMD for image retrieval

作　　者：宋和平[1] 杨群生[1] 战荫伟[1] 

机构地区：[1]广东工业大学计算机学院,广东广州510006

出　　处：《电子技术应用》2008年第3期115-118,共4页Application of Electronic Technique

基　　金：国家自然科学基金项目(60572078);广东省自然科学基金项目(05006349);武汉大学测绘遥感信息工程国家重点实验室基金项目(WKL(03)0101)

摘　　要：相似度量是图像检索的关键,EMD是一种有效的度量距离,但其计算比较复杂,而且赖于基本距离的选择。采用Lloyd聚类算法对图像进行高斯混合建模,并以聚类失真作为基本距离提出了两种近似EMD的方法计算相似度。实验结果验证了该方法的有效性,其检索效率与EMD方接近,而且计算复杂度比EMD方法低,基本距离的选择不敏感。Similarity measure is crucial for content based image retrieval, EMD is an efficient distance measurement, but it costs high computation complexity, and relies on efficient ground distance. In this paper, the images are modeled with Lloyd clus- tering, and two methods to approximate EMD with clustering distortion as ground distance are proposed. Experiment results demonstrate the effectiveness of the pro-posed methods. With similar performance compared with EMD, the proposed methods not only have lower computation complexity, but aren＇t sensitive to the ground distance.

关 键 词：图像检索 劳埃德聚类 推土机距离 最小元素法 伏格尔法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TU312.3[自动化与计算机技术—计算机科学与技术]