一种新的基于区域的图像相似性计算

为克服基于内容的图像检索(CBIR)技术中低层视觉特征无法准确地描述高层语义的问题,近年来提出的基于区域的图像检索(RBIR)技术是一种行之有效的方法。在基于区域的图像检索中,图像之间的相似性度量是影响检索结果的关键技术之一。本文提出一种新的基于区域的图像相似性的计算方法,该方法首先对图像进行区域分割,然后采用有别于传统距离度量的非线性高斯距离度量方法来计算区域对之间的相似性,实验表明该距离度量具有更好的度量性能。为了更好地接近于人对图像的理解,减小语义鸿沟,本文提出应用模糊集合理论表达区域对之间匹配的不确定性,在此基础上实现图像之间的相似性度量。最后在Corel自然图像库上对本文提出的算法进行了实验,分别与SIMPLIcity系统的IRM和UFM方法进行了比较,结果表明了本文算法的高效性。

New Region-based Image Similarity Calculation

In order to solve the problem that the low-level visual features can not express the high-level semantic features accurately in content-based image retrieval(CBIR),the region-based image retrieval technology (RBIR) has been developed in recent years as an effective method.Calculation of image similarities is one of the key technologies for good retrieval results.This paper proposes a new region-based image similarity calculating method.An image is segmented into regions,and the similarities of corresponding regions are calculated by the nonlinear Gaussian distance measuring method that is different from the traditional distance measurement.The nonlinear Gaussian distance measurement is proved a better way by experiments.To get close to the human understanding of images and reduce the semantic gap,the fuzzy set theory is presented to denote the uncertainty of choosing the matched regions in the integrated image similarities.On this basis,similarity calculation is carried out.The proposed measurement is tested in the Corel natural image database and compared with the IRM and UFM methods in the SIMIPLIcity system.The results show that the retrieval performance is greatly improved.