计算机科学与探索 ›› 2023, Vol. 17 ›› Issue (7): 1565-1575.DOI: 10.3778/j.issn.1673-9418.2210125

• 前沿·综述 • 上一篇    下一篇

深度特征的实例图像检索算法综述

季长清,王兵兵,秦静,汪祖民   

  1. 1. 大连大学 物理科学与技术学院,辽宁 大连 116622
    2. 大连大学 信息工程学院,辽宁 大连 116622
    3. 大连大学 软件学院,辽宁 大连 116622
  • 出版日期:2023-07-01 发布日期:2023-07-01

Survey of Deep Feature Instance Level Image Retrieval Algorithms

JI Changqing, WANG Bingbing, QIN Jing, WANG Zumin   

  1. 1. School of Physical Science and Technology, Dalian University, Dalian, Liaoning 116622, China
    2. School of Information Engineering, Dalian University, Dalian, Liaoning 116622, China
    3. School of Software, Dalian University, Dalian, Liaoning 116622, China
  • Online:2023-07-01 Published:2023-07-01

摘要: 基于内容的图像检索算法(CBIR)目标是在数量庞大的图像数据库中通过分析视觉内容,找出与查询图像在语义上匹配或相近的图像。其中通过特征提取获得具有判别性的图像表示对检索结果至关重要。随着深度学习的不断发展,图像检索中使用的图像特征表示方法也逐渐由原来的基于手工特征的方法转变为基于深度特征的方法。通过从特征提取的不同方法角度出发,回顾并追踪了最近基于深度特征的图像检索算法。对基于深度特征的图像检索算法分为基于深度全局特征与基于深度局部特征的图像检索算法两方面进行综述,其中在基于深度局部特征算法中重点关注了深度卷积特征聚合技术。并对现在广泛应用的深度全局与局部特征融合的图像检索方法进行归纳。探讨了深度特征的实例图像检索技术在遥感图像检索、电子商务产品检索和医疗图像检索领域中的实际应用,并比较这些特征提取算法在图像检索精度方面的表现。最后展望了深度特征提取技术在实例图像检索领域的未来研究趋势。

关键词: 实例图像检索, 深度学习, 深度全局特征, 深度局部特征

Abstract: Content-based image retrieval algorithm (CBIR) aims to find semantically matching or similar images with query images. It analyzes visual content in a large number of image databases. It is important to obtain discriminant image representation by feature extraction. With the continuous development of deep learning, the image feature representation method used in image retrieval has gradually changed. The original extraction method is  based on manual features. Now it is based on deep features. From the perspective of different feature extraction methods, the recent image retrieval algorithms based on depth feature are reviewed and traced. The image retrieval algorithms based on depth feature are divided into two aspects: depth global feature and depth local feature. The deep convolution feature aggregation technique is emphasized in the deep local feature algorithm. The widely used image retrieval methods of deep global and local feature fusion are summarized. This paper discusses the practical application of deep feature image retrieval technology in remote sensing image retrieval, e-commerce product retrieval and medical image retrieval. And it compares the performance of these feature extraction algorithms in image retrieval accuracy. Finally, the future research trend of depth feature extraction in case image retrieval is forecasted.

Key words: instance level image retrieval, deep learning, depth global feature, depth local feature