计算机科学与探索 ›› 2021, Vol. 15 ›› Issue (2): 219-232.DOI: 10.3778/j.issn.1673-9418.2007074

• 综述·探索 • 上一篇    下一篇

基于深度学习的显著性目标检测综述

史彩娟,张卫明,陈厚儒,葛录录   

  1. 华北理工大学 人工智能学院,河北 唐山 063210
  • 出版日期:2021-02-01 发布日期:2021-02-01

Survey of Salient Object Detection Based on Deep Learning

SHI Caijuan, ZHANG Weiming, CHEN Houru, GE Lulu   

  1. College of Artificial Intelligence, North China University of Science and Technology, Tangshan, Hebei 063210, China
  • Online:2021-02-01 Published:2021-02-01

摘要:

随着深度学习的不断发展,基于深度学习的显著性目标检测已经成为计算机视觉领域的一个研究热点。首先对现有的基于深度学习的显著性目标检测算法分别从边界/语义增强、全局/局部结合和辅助网络三个角度进行了分类介绍并给出了显著性图,同时对三种类型方法进行了定性分析比较;然后简单介绍了基于深度学习的显著性目标检测常用的数据集和评估准则;接着对所提基于深度学习的显著性目标检测方法在多个数据集上进行了性能比较,包括定量比较、P-R曲线和视觉比较;最后指出现有基于深度学习的显著性目标检测方法在复杂背景、小目标、实时性检测等方面的不足,并对基于深度学习的显著性目标检测的未来发展方向,如复杂背景、实时、小目标、弱监督等显著性目标检测进行了探讨。

关键词: 显著性目标检测, 深度学习, 视觉显著性

Abstract:

With the development of deep learning, salient object detection based on deep learning has become a research hotspot in the computer vision field. Firstly, existing salient object detection methods based on deep learning are introduced from three aspects, i.e. boundary/semantic enhancement, global/local combination and auxiliary network. As the same time, the saliency maps of these methods, qualitative analysis and comparison are given. Then the main datasets and main evaluation criteria for salient object detection based on deep learning are introduced in brief. Next the performance of salient object detection methods based on deep learning are compared on some datasets, including quantitative comparison, P-R curves and visual comparison. Finally, the shortcomings of the existing methods in complex background, small objects and real-time detection are pointed out, and the future development direction of salient object detection methods based on deep learning is explored, such as complex background, real-time, small object, weakly-supervised salient object detection and so on.

Key words: salient object detection, deep learning, visual saliency