《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (1): 239-244.DOI: 10.11772/j.issn.1001-9081.2021010134

• 多媒体计算与计算机仿真 • 上一篇    

基于双注意力机制信息蒸馏网络的图像超分辨率复原算法

王素玉1,2, 杨静1,2(), 李越1,2   

  1. 1.北京市物联网软件与系统工程技术研究中心(北京工业大学),北京 100124
    2.北京工业大学 信息学部,北京 100124
  • 收稿日期:2021-01-26 修回日期:2021-04-28 接受日期:2021-04-29 发布日期:2022-01-11 出版日期:2022-01-10
  • 通讯作者: 杨静
  • 作者简介:王素玉(1976—),女,河北唐山人,副教授,博士,主要研究方向:图像与视频信号处理、计算机视觉
    杨静(1994—),女,河北石家庄人,硕士研究生,主要研究方向:图像处理
    李越(1995—),男,山西阳泉人,硕士研究生,主要研究方向:图像处理。

Image super-resolution restoration algorithm based on information distillation network with dual attention mechanism

Suyu WANG1,2, Jing YANG1,2(), Yue LI1,2   

  1. 1.Beijing Engineering Research Center for IoT Software and Systems (Beijing University of Technology),Beijing 100124,China
    2.Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China
  • Received:2021-01-26 Revised:2021-04-28 Accepted:2021-04-29 Online:2022-01-11 Published:2022-01-10
  • Contact: Jing YANG
  • About author:WANG Suyu, born in 1976, Ph. D., associate professor. Her research interests include image and video signal processing, computer vision.
    YANG Jing, born in 1994, M. S. candidate. Her research interests include image processing.
    LI Yue, born in 1995, M. S. candidate. His research interests include image processing.

摘要:

针对超分辨率复原技术中网络层数不断加深导致的网络训练困难、特征信息利用率低等问题,设计并实现了一种基于双注意力的信息蒸馏网络(IDN)的图像超分辨率复原算法。首先,利用IDN较低的计算复杂度及信息蒸馏模块提取更多特征的优势,通过引入残差注意力模块(RAM)并考虑图像通道之间的相互依赖性来自适应地重新调整特征权重,从而进一步提升图像高分辨率细节的重建能力;然后,设计了对于边缘信息敏感的新型混合损失函数对图像进行细化处理,以加速网络收敛。在Set5、Set14、BSD100和Urban100公共数据集上的测试结果表明,该方法的视觉效果和峰值信噪比(PSNR)均优于当前主流算法。

关键词: 信息蒸馏网络, 图像超分辨率复原, 空间注意力, 通道注意力, 混合损失函数

Abstract:

Aiming at the problems of network training difficulty and low utilization rate of feature information caused by increasing network layers in super-resolution restoration technology, an image super-resolution restoration algorithm based on dual attention Information Distillation Network (IDN) was designed and implemented. Firstly, by taking the advantage of the low computational complexity of IDN and the advantage of the information distillation module by which more features were extracted, the weights of the features were readjust adaptively by introducing the Residual Attention Module (RAM) and considering the interdependence of image channels, so as to further improve the reconstruction ability of high-resolution details of images. Then, a new mixed loss function sensitive to edge information was designed to refine the image and accelerate the convergence of the network. Test results on Set5, Set14, BSD100 and Urban100 public datasets show that the visual effect and Peak Signal-to-Noise Ratio (PSNR) of the proposed method are superior to those of the current mainstream algorithms.

Key words: Information Distillation Network (IDN), image super-resolution restoration, Spatial Attention (SA), Channel Attention (CA), mixed loss function

中图分类号: