《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (8): 2571-2577.DOI: 10.11772/j.issn.1001-9081.2021061126

• 多媒体计算与计算机仿真 • 上一篇    

基于近似U型网络结构的图像去噪模型

靳华中, 张修洋, 叶志伟(), 张闻其, 夏小鱼   

  1. 湖北工业大学 计算机学院,武汉 430068
  • 收稿日期:2021-07-02 修回日期:2021-09-05 接受日期:2021-09-14 发布日期:2021-12-27 出版日期:2022-08-10
  • 通讯作者: 叶志伟
  • 作者简介:靳华中(1973—),男,湖北洪湖人,副教授,博士,CCF会员,主要研究方向:计算机视觉、智能系统;
    张修洋(1997—),男,湖北宜昌人,硕士研究生,主要研究方向:计算机视觉、图像处理;
    叶志伟(1978—),男,湖北浠水人,教授,博士,CCF会员,主要研究方向:机器学习、数据挖掘、进化计算;
    张闻其(2000—),男,湖北黄冈人,主要研究方向:智能计算、图像处理;
    夏小鱼(2000—),女,湖北荆州人,主要研究方向:智能计算、图像处理。

Image denoising model based on approximate U-shaped network structure

Huazhong JIN, Xiuyang ZHANG, Zhiwei YE(), Wenqi ZHANG, Xiaoyu XIA   

  1. School of Computer Science,Hubei University of Technology,Wuhan Hubei 430068,China
  • Received:2021-07-02 Revised:2021-09-05 Accepted:2021-09-14 Online:2021-12-27 Published:2022-08-10
  • Contact: Zhiwei YE
  • About author:JIN Huazhong, born in 1973, Ph. D., associate professor. His research interests include computer vision, smart system.
    ZHANG Xiuyang, born in 1997, M. S. candidate. His research interests include computer vision, image processing.
    YE Zhiwei, born in 1978, Ph. D., professor. His research interests include machine learning, data mining, evolutionary computing.
    ZHANG Wenqi, born in 2000. His research interests include intelligent computing, image processing.
    XIA Xiaoyu, born in 2000. Her research interests include intelligent computing, image processing.
  • Supported by:
    National College Students’ Innovation and Entrepreneurship Training Program(202010500003)

摘要:

针对图像去噪中的去噪效果差、训练周期长的问题,提出一种基于近似U型网络结构的图像去噪模型。首先,使用不同步长的卷积层将原有的线性网络结构修改为近似U型的网络结构;然后,将不同感受野的图像信息叠加以尽可能地保留图像的原有信息;最后,引入反卷积网络层进行图像恢复和噪声的进一步去除。在Set12与BSD68测试集上与去噪卷积神经网络(DnCNN)模型相比,所提模型的峰值信噪比(PSNR)平均提升了0.04~0.14 dB,训练时长平均缩短了41%。实验结果表明,所提模型具有更好地去噪效果和更短的训练时长。

关键词: 图像去噪, 去噪卷积神经网络, 反卷积, U-Net, 残差学习

Abstract:

Aiming at the problem of poor denoising effect and long training period in image denoising, an image denoising model based on approximate U-shaped network structure was proposed. Firstly, the original linear network structure was modified to an approximate U-shaped network structure by using convolutional layers with different strides. Then, the image information of different receptive fields was superimposed on each other to preserve the original information of the image as much as possible. Finally, the deconvolutional network layer was introduced for image restoration and further noise removal. Experimental results show that on Set12 and BSD68 test sets: compared with Denoising Convolutional Neural Network (DnCNN) model, the proposed model has an average increase of 0.04 to 0.14 dB on Peak Signal-to-Noise Ratio (PSNR), and an average reduction of 41% on training time, verifying that the proposed model has better denoising effect and shorter training time.

Key words: image denoising, Denoising Convolutional Neural Network (DnCNN), deconvolution, U-Net, residual learning

中图分类号: