Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (8): 28-40.DOI: 10.3778/j.issn.1002-8331.2208-0223

• Research Hotspots and Reviews • Previous Articles     Next Articles

Review of Research on Real-World Single Image Super-Resolution Reconstruction

ZHANG Yanqing, MA Jianhong, HAN Ying, CAO Yangjie, LI Jie, YANG Cong   

  1. 1.School of Cyber Science and Technology, Zhengzhou University, Zhengzhou 450002, China
    2.Department of Computer Science and Engineering, Shanghai Jiaotong University, Shanghai 200030, China
  • Online:2023-04-15 Published:2023-04-15

真实场景下图像超分辨率重建研究综述

张艳青,马建红,韩颖,曹仰杰,李颉,杨聪   

  1. 1.郑州大学 网络空间安全学院,郑州 450002
    2.上海交通大学 计算机科学与工程系,上海 200030

Abstract: Single image super-resolution is an important research topic in the field of computer vision in recent decades. The super-resolution reconstruction algorithm based on deep learning has made breakthroughs. When the super-resolution algorithm is applied to the image in real scene, the effect will be greatly reduced, and serious blur and ringing effect will appear. In this context, more and more researchers are committed to the study of real-world single image SR(RSISR) algorithm. Taking RSISR as the research object, this paper first introduces common image data sets and evaluation indexes, and then analyzes and compares the characteristics, performance and shortcomings of various methods from two aspects:SR methods based on external data sets and SR methods based on internal data sets. Finally, the difficulties and challenges of RSISR are discussed, and the future development trend is considered and prospected.

Key words: single image super-resolution, real scene, deep learning, super-resolution data set

摘要: 单幅图像超分辨率是近几十年来计算机视觉领域的一个重要研究课题,基于深度学习的超分辨率重建算法已经取得突破性进展,但当大多数算法应用到真实场景中的图像时效果会大大降低,出现严重模糊、振铃效应等。在此背景下越来越多研究人员致力于研究真实场景下的图像超分辨率算法(real-world single image super-resolution,RSISR)。聚焦于真实场景下图像超分辨率重建算法,介绍了常用公共图像数据集和图像评估指标,从基于外部数据集SR方法和基于内部数据集SR方法两大方向分析对比了各种方法的特点、性能和不足。讨论了RSISR存在的困难和挑战,并对未来的发展趋势进行了思考与展望。

关键词: 单幅图像超分辨率, 真实场景, 深度学习, 超分辨率数据集