Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 62, Issue 2, 0237010(2025)

Single-Image Super-Resolution Reconstruction Based on Improved Attention in A2N

Hualiang Cao* and Wei Zhuang
Author Affiliations
  • Jiangsu Key Laboratory of Opto-Electronic Technology, School of Physics and Technology, Nanjing Normal University, Nanjing 210023, Jiangsu , China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (16)
    Equations (4)
    References (32)
    Tools
    Paper Information
    Topics
    Figures & Tables(16)
    Schematic diagram of the overall structure of A2N[10]

    Fig. 1. Schematic diagram of the overall structure of A2N[10]

    Download full size
    Architecture of A2B[10]

    Fig. 2. Architecture of A2B[10]

    Download full size
    Schematic diagram of the overall structure of RFB-A2N

    Fig. 3. Schematic diagram of the overall structure of RFB-A2N

    Download full size
    RFB. (a) Architecture of RFB; (b) channel number setting in RFB

    Fig. 4. RFB. (a) Architecture of RFB; (b) channel number setting in RFB

    Download full size
    Schematic diagrams of a gridding artifact

    Fig. 5. Schematic diagrams of a gridding artifact

    Download full size
    Architecture of image reconstruction module

    Fig. 6. Architecture of image reconstruction module

    Download full size
    Comparison in FLOPs, runtime, and number of parameters

    Fig. 7. Comparison in FLOPs, runtime, and number of parameters

    Download full size
    Comparison of ×2 reconstruction results for image 58060 in the B100 dataset

    Fig. 8. Comparison of ×2 reconstruction results for image 58060 in the B100 dataset

    Download full size
    Comparison of ×2 reconstruction results for image ppt3 in the Set14 dataset

    Fig. 9. Comparison of ×2 reconstruction results for image ppt3 in the Set14 dataset

    Download full size
    Comparison of ×4 reconstruction results for image 148026 in the B100 dataset

    Fig. 10. Comparison of ×4 reconstruction results for image 148026 in the B100 dataset

    Download full size
    Comparison of ×4 reconstruction results for image img_19 in the Urban100 dataset

    Fig. 11. Comparison of ×4 reconstruction results for image img_19 in the Urban100 dataset

    Download full size

    • Table 1. Comparison of PSNR of two different feature fusion methods on the benchmark dataset

      View table

      Table 1. Comparison of PSNR of two different feature fusion methods on the benchmark dataset

      DatasetA2N-M-B10AddConcat
      Set532.21732.17632.164
      Set1428.60328.60828.610
      B10027.56627.57727.576
      Urban10026.13626.11226.147
      Manga10930.45330.47730.538

    • Table 2. Impact of RFB on network PSNR

      View table

      Table 2. Impact of RFB on network PSNR

      BlockHFFResidualParams /106FLOPs /109Set5Set14B100Urban100Manga109
      A2B0.5348.6532.16428.61027.57626.14730.538
      RFB-A2B√0.6355.1632.31728.64827.59926.24730.621
      RFB-A2B√0.6355.1632.23928.68927.62126.27230.650
      RFB-A2B√√0.6355.1632.28728.68927.62526.27130.700
      A方正汇总行2N-M（baseline）0.8069.1432.27028.70027.61026.28030.590

    • Table 3. Effect of different loss functions on network PSNR

      View table

      Table 3. Effect of different loss functions on network PSNR

      ModelLoss functionSet5Set14B100Urban100Manga109
      A2N-ML1 Loss32.2728.7027.6126.2830.59
      RFB-A2NL1 Loss32.28728.68927.62526.27130.700
      RFB-A2NCombined loss32.28228.74327.65526.36030.830

    • Table 4. Impact of the number of RFB-A2B on network performance

      View table

      Table 4. Impact of the number of RFB-A2B on network performance

      Number of blocksParams /106FLOPs /109Runtime /sSet5Set14B100Urban100Manga109
      80.547.340.14232.22828.67527.63926.28330.788
      100.655.160.16832.28228.74327.65526.36030.830
      120.762.980.19232.29928.76027.68126.42630.890
      A2N-M0.869.140.19732.2728.7027.6126.2830.59
      A2N182.560.21132.3028.7127.6126.2730.67

    • Table 5. Comparison of PSNR and SSIM values of different algorithms at ×2 and ×4 upscale factors

      View table

      Table 5. Comparison of PSNR and SSIM values of different algorithms at ×2 and ×4 upscale factors

      AlgorithmParams /106

      Set5

      PSNR/SSIM

      Set14

      PSNR/SSIM

      B100

      PSNR/SSIM

      Urban100

      PSNR/SSIM

      Manga109

      PSNR/SSIM

      upscale： ×2
      SRCNN0.0636.66/0.954232.45/0.906731.36/0.887929.50/0.894635.60/0.9663
      VDSR0.737.53/0.958733.03/0.912431.90/0.896030.76/0.914037.22/0.9729
      IMDN0.738.00/0.960533.63/0.917732.19/0.899632.17/0.928338.88/.9774
      RFDN0.538.05/0.960633.68/0.918432.16/0.899432.12/0.927838.88/0.9773
      RLFN0.538.07/0.960733.72/0.918732.22/0.900032.33/0.9299
      PAN0.338.00/0.960533.59/0.918132.18/0.899732.01/0.927338.70/0.9773
      RepRFN0.437.99/0.960933.57/0.917932.18/0.900431.95/0.926138.80/0.9774
      AWSRN-M138.04/0.960533.66/0.918132.21/0.900032.23/0.929438.66/0.9772
      CARN1.637.76/0.959033.52/0.916632.09/0.897831.92/0.9256
      A2N138.06/0.960833.75/0.919432.22/0.900232.43/0.931138.87/0.9769
      A2N-M0.838.06/0.960133.73/0.919032.22/0.899732.34/0.930038.80/0.9765
      RFB-A2N0.638.12/0.961033.73/0.919432.24/0.900632.39/0.930739.00/0.9775
      AlgorithmParams /106

      Set5

      PSNR/SSIM

      Set14

      PSNR/SSIM

      B100

      PSNR/SSIM

      Urban100

      PSNR/SSIM

      Manga109

      PSNR/SSIM

      upscale： ×4
      SRCNN0.0630.48/0.862827.49/0.750326.90/0.710124.52/0.722127.66/0.8505
      VDSR0.731.35/0.883828.01/0.767427.29/0.725125.18/0.752428.83/0.8809
      IMDN0.732.21/0.894828.58/0.781127.56/0.735326.04/0.783830.45/0.9075
      RFDN0.532.24/0.895228.61/0.781927.57/0.736026.11/0.785830.58/0.9089
      RLFN0.532.24/0.895228.62.0.781327.60/0.736426.17/0.7877
      PAN0.332.13/0.894828.61/0.782227.59/0.736326.11/0.785430.51/0.9095
      RepRFN0.432.15/0.895228.63/0.782427.60/0.737726.09/.0.783430.52/0.9075
      AWSRN-M1.332.21/0.895428.65/0.783227.60/0.736826.15/0.788430.56/0.9093
      CARN1.632.13/0.893728.60/0.780627.58/0.734926.07/0.7837
      A2N132.30/0.896628.71/0.784227.61/0.737426.27/0.792030.67/0.9110
      A2N-M0.832.27/0.896328.70/0.784227.61/0.737626.28/0.791930.59/0.9103
      RFB-A2N0.632.28/0.895828.74/0.785227.66/0.738426.36/0.792930.83/0.9120
    Tools

    Get Citation

    Copy Citation Text

    Hualiang Cao, Wei Zhuang. Single-Image Super-Resolution Reconstruction Based on Improved Attention in A2N[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0237010

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Digital Image Processing

    Received: Apr. 29, 2024

    Accepted: Jun. 3, 2024

    Published Online: Jan. 6, 2025

    The Author Email:

    DOI:10.3788/LOP241193

    CSTR:32186.14.LOP241193

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology