计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (1): 196-202.DOI: 10.3778/j.issn.1002-8331.1709-0209

• 图形图像处理 • 上一篇    下一篇

基于文本与语义相关性分析的图像检索

穆亚昆,冯圣威,张  静   

  1. 华东理工大学 信息科学与工程学院,上海 200237
  • 出版日期:2019-01-01 发布日期:2019-01-07

Image Retrieval Based on Text and Semantic Relevance Analysis

MU Yakun, FENG Shengwei, ZHANG Jing   

  1. College of Information Science and Engineering, East China University of Science and Technology, Shanghai 200237, China
  • Online:2019-01-01 Published:2019-01-07

摘要: 为了更加有效地检索到符合用户复杂语义需求的图像,提出一种基于文本描述与语义相关性分析的图像检索算法。该方法将图像检索分为两步:基于文本语义相关性分析的图像检索和基于SIFT特征的相似图像扩展检索。根据自然语言处理技术分析得到用户文本需求中的关键词及其语义关联,在选定图像库中通过语义相关性分析得到“种子”图像;接下来在图像扩展检索中,采用基于SIFT特征的相似图像检索,利用之前得到的“种子”图像作为查询条件,在网络图像库中进行扩展检索,并在结果集上根据两次检索的图像相似度进行排序输出,最终得到更加丰富有效的图像检索结果。为了证明算法的有效性,在标准数据集Corel5K和网络数据集Deriantart8K上完成了多组实验,实验结果证明该方法能够得到较为精确地符合用户语义要求的图像检索结果,并且通过扩展算法可以得到更加丰富的检索结果。

关键词: 图像检索, 基于文本语义相关性的图像检索, 语义相关度, SIFT低层视觉特征, 图像扩展检索

Abstract: An image retrieval algorithm based on text and semantic relevance is proposed to effectively retrieve images, which can satisfy the complex requirement of users. There are two steps in this method:images retrieval based on textual semantic relevance analysis and image extension retrieval based on SIFT features. Firstly, according to the analysis of Natural Language Processing technology, it acquires text keywords and semantic correlation of users’ demands, and applies them to retrieve seed images from the image dataset by semantic relevance analysis. Then image extension retrieval based on SIFT feature are used to conduct extend retrieval in Web image datasets according to the retrieval results of last step. Finally, the paper achieves the query results by combining the retrieval results of these two steps. The experiments on Standard Data Set Corel5K and Web Data Set Deriantart8K prove that this method can achieve more precise image retrieval results by semantic relevance analysis and can enrich the query results by extend retrieval.

Key words: image retrieval, image retrieval based on semantic correlation of text, semantic relevance, low-rise visual features of SIFT, image extension retrieval