多级信息补偿的U型网络图像超分辨率重建算法
DOI:
作者:
作者单位:

兰州理工大学电气工程与信息工程学院

作者简介:

通讯作者:

中图分类号:

TP391

基金项目:

国家自然科学基金(61763029),国家重点研发计划 (2020YFB1713600),国防基础科研项目(JCKY2018427C002),甘肃省教育厅产业支撑计划项目(2021CYZC-02),甘肃省科技计划资助(21JR7RA206),甘肃省科技计划资助(21YF5GA072),甘肃省自然科学基金(20JR5RA459).


Super-resolution reconstruction algorithm of U-shaped network image based on multi-level information compensation
Author:
Affiliation:

College of Electrical Engineering and Information Engineering, Lanzhou University of Technology

Fund Project:

the National Natural Science Foundation of China (No.61763029)the Science and Technology Project of Gansu Province(21YF5GA072)the Science and Technology Project of Gansu Province(21JR7RA206) the National Key Research and Development Plan (2020YFB1713600)Industrial Support Project of Education Department of Gansu Province(2021CYZC-02)the National Defense Basic Research Project of China (JCKY2018427C002)the Natural Science Foundation of Gansu(20JR5RA459)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对基于深度神经网络的图像超分辨率重建算法在特征提取过程中容易丢失特征信息,导致重建图像缺少纹理和边缘细节等问题,本文提出了一种多级信息补偿的U型网络图像超分辨率重建算法。首先设计了一个用于图像超分辨率重建的U型网络,该网络通过下通道分支对输入特征进行多层级特征提取和通道压缩,通过底层模块对压缩后的特征进行融合并提取不同通道的相关特征和通过上通道分支对压缩后的相关特征进行多层次特征提取和通道恢复;然后设计了多级信息补偿模型,该方法对U型网络的通道压缩过程中丢失的信息和通道恢复过程中难以恢复的信息进行补偿。最后在不同放大倍数下的Set5、Set14、BSD100和Urban100测试集上对本文算法和主流算法进行对比测试分析,实验结果表明本文算法相比主流算法不仅实现了在峰值信噪比(PSNR)/结构相似度(SSIM)指标和视觉效果的巨大提升,而且拥有较少的参数量。

    Abstract:

    Aiming at the problem that the image reconstruction algorithm based on deep neural network is easy to lose feature information during feature extraction, which leads to the lacks of texture and edge details in the reconstructed image, this paper proposes a U-shaped network image reconstruction algorithm with multi-level information compensation. Firstly, a U-shaped network for image reconstruction is designed, which performs multi-level feature extraction and channel compression on the input features through the lower channel branch, and fuses the compressed features through the bottom module and extracts the related features of different channels, and performs multi-level feature extraction and channel recovery on the compressed related features through the upper channel branch. Then, a multi-level information compensation model is designed, which compensates for the lost information in the channel compression process and the information that is difficult to recover in the channel recovery process of U-shaped network. Finally, compared with the mainstream algorithms, the proposed algorithm is tested and analyzed by the test sets of Set5, Set14, BSD100, and Urban100 at different magnifications. The experimental results show that the proposed algorithm not only greatly improves the peak signal-to-noise ratio (PSNR)/structural similarity (SSIM) index and visual effect, but also has fewer parameters.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2021-12-27
  • 最后修改日期:2022-04-21
  • 录用日期:2022-04-27
  • 在线发布日期:
  • 出版日期: