摘要： 基于相似性度量的图像检索方法大多仅考虑检索图像与结果图像之间的距离，而不考虑结果之间的关系，使得检索精度受到影响。该文提出了基于加权关系图谱特征的图像检索算法，该算法利用检索图像与检索初始结果图像的距离构造加权关系图，利用该关系图的谱系数夹角特征确定最终输出的检索结果。对比检索实验表明，该方法可以提高检索的精度、具有较好的稳定性。

关键词: 基于内容的图像检索, 加权关系图, 图谱特征

Abstract: Most content-based image retrieval algorithms intend to get results images similar to the query image. But few of them cares for the similarity among the results. To solve this problem, a novel algorithm based on attributed relational graph is presented for image retrieval. This algorithm builds attributed relational graph on extended initial results. A new spectral feature is extracted from the graph. It is called angle between spectral coefficient vectors. Then the new feature is employed to guarantee the similarity of final result images. The experimental results show that this method can improve the retrieval accuracy than other existing methods. And the algorithm does not increase the complexity of the algorithm evidently.

Key words: content-based image retrieval, attributed relational graph, graph spectral feature

中图分类号: 

• TP391.41