Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (23): 238-244.DOI: 10.3778/j.issn.1002-8331.2105-0325

• Graphics and Image Processing • Previous Articles     Next Articles

Super-Resolution of Fundus Images by Information Distillation and Heterogeneous Up-Sampling

LI Hangyu, XUAN Zuxing, ZHOU Jianping, HU Xiyuan, CHENG Gangwei   

  1. 1.Beijing Key Laboratory of Information Service Engineering, Beijing Union University, Beijing 100101, China
    2.Institute of Fundamental and Interdisciplinary Sciences, Beijing Union University, Beijing 100101, China
    3.School of Computer Science and Technology, Anhui University of Technology, Ma’anshan, Anhui 243032, China
    4.School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
    5.Ophthalmic Cenrter, Peking Union Medical College Hospital, Beijing 100005, China
  • Online:2022-12-01 Published:2022-12-01

信息蒸馏与异构上采样的眼底图像超分辨重建

李航宇,玄祖兴,周建平,胡晰远,程钢炜   

  1. 1.北京联合大学 北京市信息服务工程重点实验室,北京 100101
    2.北京联合大学 数理与交叉科学研究院,北京 100101
    3.安徽工业大学 计算机科学与技术学院,安徽 马鞍山 243032
    4.南京理工大学 计算机科学与工程学院,南京 210094
    5.北京协和医院 眼科中心,北京 100005

Abstract: Large-scale disease screening in poor areas relies on portable hand-held cameras and remote diagnostics, which often produce poor quality and low-resolution images of the fundus. To solve this problem, a lightweight super-resolution network based on information distillation and heterogeneous up-sampling is designed. Considering the difference between fundus images and natural images, the network makes use of distillation features to supplement coarse features, then up samples coarse features and depth features in a heterogeneous way, and finally integrates two kinds of up sampled features to obtain high-definition fundus images. Compared with the advanced method from three aspects of image quality, parameter memory and run-time, the highest image quality is achieved while the parameter memory and run-time are excellent, which provides an idea for super-resolution algorithm to be embedded in handheld fundus cameras and general medical devices.

Key words: fundus image, super-resolution, information distillation, heterogeneous up-sampling

摘要: 贫困地区的大规模疾病筛查主要依靠便携式手持摄像机和远程诊断来完成,这一过程获得的眼底图像往往质量较差,分辨率较低。针对这一问题,设计了一种基于信息蒸馏与异构上采样的轻量级超分辨网络。网络考虑到眼底图像与自然图像的区别,利用蒸馏特征对粗特征进行补足,然后以异构的方式将粗特征与深度特征分别进行上采样,最后集成两种上采样的特征得到高清眼底图像。从图像质量、参数内存和运行时间三个方面与先进的方法进行比较,在参数内存和运行时间成绩优异的同时取得了最高的图像质量,这为超分辨算法嵌入在手持眼底摄像机和普通医用设备上提供了思路。

关键词: 眼底图像, 超分辨率, 信息蒸馏, 异构上采样