《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (5): 1579-1587.DOI: 10.11772/j.issn.1001-9081.2023050689

• 多媒体计算与计算机仿真 • 上一篇    

基于感受野扩展残差注意力网络的图像超分辨率重建

郭琳1,2,3, 刘坤虎1(), 马晨阳1, 来佑雪1, 徐映芬1   

  1. 1.湖北大学 人工智能学院, 武汉 430062
    2.智能感知系统与安全教育部重点实验室, 武汉 430062
    3.智慧政务与人工智能应用湖北省工程研究中心, 武汉 430062
  • 收稿日期:2023-06-01 修回日期:2023-09-01 接受日期:2023-09-12 发布日期:2023-09-14 出版日期:2024-05-10
  • 通讯作者: 刘坤虎
  • 作者简介:郭琳(1978—),女,湖北随州人,副教授,博士,主要研究方向:信号处理、机器视觉、深度学习
    马晨阳(1998—),男,河南驻马店人,硕士研究生,主要研究方向:深度学习
    来佑雪(2000—),女,山东济宁人,硕士研究生,主要研究方向:图像处理
    徐映芬(2003—),女,安徽安庆人,主要研究方向:计算机视觉。
    第一联系人:刘坤虎(1998—),男,湖北荆州人,硕士研究生,主要研究方向:图像超分辨率重建
  • 基金资助:
    国家自然科学基金资助项目(62273135)

Image super-resolution reconstruction based on residual attention network with receptive field expansion

Lin GUO1,2,3, Kunhu LIU1(), Chenyang MA1, Youxue LAI1, Yingfen XU1   

  1. 1.School of Artificial Intelligence,Hubei University,Wuhan Hubei 430062,China
    2.Key Laboratory of Intelligent Perception Systems and Security of Ministry of Education,Wuhan Hubei 430062,China
    3.Hubei Provincial Engineering Research Center for Smart Government Affairs and Artificial Intelligence Application,Wuhan Hubei 430062,China
  • Received:2023-06-01 Revised:2023-09-01 Accepted:2023-09-12 Online:2023-09-14 Published:2024-05-10
  • Contact: Kunhu LIU
  • About author:GUO Lin, born in 1978, Ph. D., associate professor. Her research interests include signal processing, machine vision, deep learning.
    MA Chenyang, born in 1998, M. S. candidate. His research interests include deep learning.
    LAI Youxue, born in 2000, M. S. candidate. Her research interests include image processing.
    XU Yingfen, born in 2003. Her research interests include computer vision.
  • Supported by:
    National Natural Science Foundation of China(62273135)

摘要:

针对现有残差网络存在残差特征利用不充分、细节丢失的问题,提出一种结合两层残差聚合结构和感受野扩展双注意力机制的深度神经网络模型,用于单幅图像超分辨率(SISR)重建。该模型通过跳跃连接形成两层嵌套的残差聚合网络结构,对网络各层提取的大量残差信息进行分层聚集和融合,能减少包含图像细节的残差信息的丢失。同时,设计一种多尺度感受野扩展模块,能捕获更大范围、不同尺度的上下文相关信息,促进深层残差特征的有效提取;并引入空间-通道双注意力机制,增强残差网络的判别性学习能力,提高重建图像质量。在数据集Set5、Set14、BSD100和Urban100上进行重建实验,并从客观指标和主观视觉效果上将所提模型与主流模型进行比较。客观评价结果表明,所提模型在全部4个测试数据集上均优于对比模型,其中,相较于经典的超分辨率卷积神经网络(SRCNN)模型和性能次优的对比模型ISRN(Iterative Super-Resolution Network),在放大2倍、3倍、4倍时的平均峰值信噪比(PSNR)分别提升1.91、1.71、1.61 dB和0.06、0.04、0.04 dB;视觉效果对比显示,所提模型恢复的图像细节纹理更清晰。

关键词: 图像超分辨率, 残差网络, 感受野, 深度学习, 注意力

Abstract:

To solve the problems of insufficient utilization of residual features and loss of details in existing residual networks, a deep neural network model combining the two-layer structure of residual aggregation and dual-attention mechanism with receptive field expansion, was proposed for Single Image Super-Resolution (SISR) reconstruction. In this model, a two-layer nested network structure of residual aggregation was constructed through skip connections, to agglomerate and fuse hierarchically the residual information extracted by each layer of the network, thereby reducing the loss of residual information containing image details. Meanwhile, a multi-scale receptive field expansion module was designed to capture a larger range of context-dependent information at different scales for the effective extraction of deep residual features; and a space-channel dual attention mechanism was introduced to enhance the discriminative learning ability of the residual network, thus improving the quality of reconstructed images. Quantitative and qualitative assessments were performed on benchmark datasets Set5, Set14, B100 and Urban100 for comparison with the mainstream methods. The objective evaluation results indicate that the proposed method outperforms the comparative methods on all four datasets; compared with the classical SRCNN (Super-Resolution using Convolutional Neural Network) model and second best performing comparison model ISRN (Iterative Super-Resolution Network), the proposed model improves the average values of Peak Signal-to-Noise Ratio (PSNR) by 1.91, 1.71, 1.61 dB and 0.06, 0.04, 0.04 dB, respectively, at the magnification of 2, 3 and 4. Visual effects show that the proposed model reconstructs clearer image details and textures.

Key words: image super-resolution, residual network, receptive field, deep learning, attention

中图分类号: