Abstract: How to retrieve similar images from large images database efficiently is a great challenge for content-based image retrieval system. A modified K-means algorithm was proposed to form the hierarchy of the indexing structure. A* search algorithm, triangle inequality principle and N-near neighbours were applied to achieve an optimal search in order to retrieve efficiently for the large image database. Experiments on Corel database show that the proposed algorithm achieves efficient logarithm retrieval.

Key words: K-means cluster algorithm, A* tree algorithm, triangle inequality principal, N-near neighbours method

摘要： 对于大型图像库，如何高效地检索出相似图像是图像检索系统的一大挑战。提出了一种改进的K-均值聚类算法建立分层结构的索引，再利用A*树算法和三角不等式原则及N近邻方法对索引库快速高效地搜索，达到对图像库快速高效检索相似图像的目的。实验在Corel图像库上进行，实验结果表明该方法以对数时间复杂度实现基于内容的高效检索。

关键词: K-均值聚类算法, A*树算法, 三角不等式原则, N近邻方法