CATIRI: An Efficient Method for Content-and-Text Based Image Retrieval
CATIRI: An Efficient Method for Content-and-Text Based Image Retrieval
复制标题
CATIRI：一种基于内容和文本的图像检索的有效方法
DOI：
10.1007/s11390-019-1911-2
复制
发表时间：
2019
影响因子：
0.7
通讯作者：
Minyi Guo
中科院分区：
文献类型：
--
作者：
Mengqi Zeng;Bin Yao;Zhi-Jie Wang;Yanyan Shen;Feifei Li;Jianfeng Zhang;Hao Lin;Minyi Guo
图像检索中视觉和文本信息的结合显著地缓解了传统图像检索方法的语义鸿沟，因此近年来受到了广泛的关注。基于这种组合的图像检索通常被称为基于内容和文本的图像检索(CTBIR)。然而，现有的CTBIR研究主要集中在提高检索质量上。就我们所知，人们对如何提高检索效率的关注还很少。如今，图像数据在我们的日常生活中广泛存在并迅速膨胀。显然，研究检索效率是非常重要和有趣的。为此，本文提出了一种高效的图像检索方法CATIRI(Content-and-Text Based Image Retrivery Using Indeding)。CATIRI遵循一个三阶段解决方案框架，该框架开发了一种名为MHIM-TREE的新索引结构。MHIM-树无缝集成了几个元素，包括曼哈顿散列、倒排索引和M-树。为了在查询中明智地使用我们的MHIM树，我们提供了一组重要的度量并揭示了它们的固有属性。在此基础上，提出了一种面向CTBIR的top-k查询算法。基于基准图像数据集的实验结果表明，CATIRI的性能比竞争对手高出一个数量级。
The combination of visual and textual information in image retrieval remarkably alleviates the semantic gap of traditional image retrieval methods, and thus it has attracted much attention recently. Image retrieval based on such a combination is usually called the content-and-text based image retrieval (CTBIR). Nevertheless, existing studies in CTBIR mainly make efforts on improving the retrieval quality. To the best of our knowledge, little attention has been focused on how to enhance the retrieval efficiency. Nowadays, image data is widespread and expanding rapidly in our daily life. Obviously, it is important and interesting to investigate the retrieval efficiency. To this end, this paper presents an efficient image retrieval method named CATIRI (content-and-text based image retrieval using indexing). CATIRI follows a three-phase solution framework that develops a new indexing structure called MHIM-tree. The MHIM-tree seamlessly integrates several elements including Manhattan Hashing, Inverted index, and M-tree. To use our MHIM-tree wisely in the query, we present a set of important metrics and reveal their inherent properties. Based on them, we develop a top-k query algorithm for CTBIR. Experimental results based on benchmark image datasets demonstrate that CATIRI outperforms the competitors by an order of magnitude.