图像索引和相关反馈在RBIR中的应用

Application of image indexing and relevance feedback in RBIR

图像索引和相关反馈是改进基于区域的图像检索(RBIR)的2种有效的方法.本研究在图像索引方面采用随机局部搜索(RLS)算法对图像上的区域进行聚类,并引入倒查文件技术对图像库进行索引;在相关反馈方面,采用基于支持向量机的相关反馈算法来改善检索结果,并提出了一种新的核函数使该算法更适宜基于区域的图像表示.在海量图像数据库上的实验结果说明了该算法的有效性.

Image indexing and relevance feedback are two effective ways to improve region - based image retrieval （RBIR）. This paper will use randomised local search （RLS） algorithm to cluster the image regions. And then, an indexing scheme similar to the inverted file technology will be introduced. We also propose a support vector machine - based relevance feedback algorithms to refine the retrieval result. A new kernel function is proposed so as to make the algorithm be applicable for region - based representations. Experiment results on large image database demonstrate the effectiveness of the proposed algorithm.