loading page

A Deep Learning Architecture for Blind Image Super-Resolution in Medical Image System
  • +3
  • Yinghua Li,
  • Yue Liu,
  • Jian Xu,
  • Dan Xu,
  • Shengchuan Zhang,
  • Ying Liu
Yinghua Li
Xi'an University of Posts and Telecommunications
Author Profile
Yue Liu
Xi'an University of Posts and Telecommunications
Author Profile
Jian Xu
Xi'an University of Posts and Telecommunications
Author Profile
Dan Xu
Sichuan Jiuzhou Electric Group Co Ltd
Author Profile
Shengchuan Zhang
Xiamen University School of Informatics

Corresponding Author:[email protected]

Author Profile
Ying Liu
Xi'an University of Posts and Telecommunications
Author Profile

Abstract

Currently, the majority of single image super-resolution algorithms based on convolutional neural networks presume that image degradation is always bicubic down sampling. In practice, however, the mechanism of medical image degradation is considerably more complicated than the Artificially simulated degradation. When the medical image degradation differs from the presumption, this algorithm performs significantly worse than anticipated. we use a degradation comparative learning to solve such problems. In addition, many CNN-based methods currently treat various channels equally, despite the fact that their information content is not equal, resulting in the underutilization of a great deal of information. As a result, the reconstructed SR pictures' texture features are not sufficiently rich. In order to solve these issues, this paper suggests a blind image super-resolution approach based on edge reconstruction and an image feature supplement module. This method reconstructs high-resolution images with rich texture information from low-resolution images and accomplishes excellent performance in the stable restoration of various degraded SR images. By comparing our approach to other blind SR approaches using natural picture testing datasets and medical images, we were able to show that it outperformed them. The results of the experiments demonstrate that our method is superior to the most cutting-edge ones currently in use.