Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (15): 18-36.DOI: 10.3778/j.issn.1002-8331.2203-0312

• Research Hotspots and Reviews • Previous Articles     Next Articles

Review of Cross-Domain Image Retrieval

LI Haoran, ZHOU Xiaoping, WANG Jia   

  1. School of Electrical and Information Engineering, Beijing University of Civil Engineering and Architecture, Beijing 100044, China
  • Online:2022-08-01 Published:2022-08-01

跨域图像检索综述

李浩然,周小平,王佳   

  1. 北京建筑大学 电气与信息工程学院,北京 100044

Abstract: Due to the influence of imaging carriers, imaging spectra and imaging conditions, the application of cross-domain images in different fields is increasing, and cross-domain image retrieval has become a hot topic and a preamble of research in many fields. However, cross-domain image retrieval faces the problem of image visual deviation, and the results cannot be effectively obtained through traditional same-domain image retrieval methods. Through literature research, systematically reviews the representative approaches in the field of cross-domain image retrieval in recent years. The task of cross-domain image retrieval is described and the key issues are pointed out. Cross-domain image retrieval methods are divided into two categories according to different transformation stages of image domains:cross-domain image retrieval methods based on feature space migration and cross-domain image retrieval methods based on image domain adaptation, and the two types of methods are summarised and analyzed systematically, and then, the datasets of cross-domain image retrieval in different fields are collated, and the performance of various methods is compared. The existing cross-domain retrieval methods are summarised and the future research directions are outlined.

Key words: cross-domain image retrieval, deep learning, feature space migration, image domain adaptation

摘要: 受成像载体、成像光谱和成像条件等的影响,跨域图像在不同领域的应用日益增多,跨域图像检索已成为了许多领域研究的热点和前言。然而图像的跨域检索面临着图像视觉偏差的问题,通过传统同域图像检索的方法无法有效地得到结果。通过文献调研,系统梳理了近年来跨域图像检索领域的代表性方法。对跨域图像检索任务作出了简要说明并指出了关键问题;根据图像域的不同转换阶段,将跨域图像检索方法分为两类:基于特征空间迁移和基于图像域迁移的跨域图像检索方法,并对两类方法进行了系统总结和分析;整理了跨域图像检索在不同领域的数据集,对比了各类方法的性能;总结了现有跨域检索方法并对未来的研究方向进行了展望。

关键词: 跨域图像检索, 深度学习, 特征空间迁移, 图像域迁移