• Laser & Optoelectronics Progress
  • Vol. 60, Issue 10, 1010017 (2023)
Yanfei Peng, Manting Zhang*, Pingjia Zhang, Jian Li, and Lirui Gu
Author Affiliations
  • School of Electronic and Information Engineering, Liaoning Technical University, Huludao 125105, Liaoning , China
  • show less
    DOI: 10.3788/LOP220752 Cite this Article Set citation alerts
    Yanfei Peng, Manting Zhang, Pingjia Zhang, Jian Li, Lirui Gu. Single-Image Super-Resolution Reconstruction Aggregating Residual Attention Network[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010017 Copy Citation Text show less
    Comparison of ResNet and ResNeXt. (a) ResNet structure; (b) ResNeXt structure; (c) proposed residual structure
    Fig. 1. Comparison of ResNet and ResNeXt. (a) ResNet structure; (b) ResNeXt structure; (c) proposed residual structure
    Attention mechanism
    Fig. 2. Attention mechanism
    Flowchart of improved network
    Fig. 3. Flowchart of improved network
    Generator network structure
    Fig. 4. Generator network structure
    Discriminator network structure
    Fig. 5. Discriminator network structure
    Effect of cardinality number on module performance
    Fig. 6. Effect of cardinality number on module performance
    Experimental comparison result for image on BSD100 dataset. (a) Original image; (b) ResNeXt; (c) ResNeXt+SE; (d) ResNeXt+SimAM
    Fig. 7. Experimental comparison result for image on BSD100 dataset. (a) Original image; (b) ResNeXt; (c) ResNeXt+SE; (d) ResNeXt+SimAM
    Comparison of reconstruction effect of "baby" on Set5 dataset
    Fig. 8. Comparison of reconstruction effect of "baby" on Set5 dataset
    Comparison of reconstruction effect of “foreman” on Set14 dataset
    Fig. 9. Comparison of reconstruction effect of “foreman” on Set14 dataset
    Comparison of reconstruction effect of “3096” on BSD100 dataset
    Fig. 10. Comparison of reconstruction effect of “3096” on BSD100 dataset
    MethodPSNR /dBSSIM
    Baseline27.330.7517
    Baseline +ResNeXt27.830.7767
    Baseline +SimAM27.520.7651
    Baseline +SN27.570.7763
    Baseline+Charbonnier27.440.7672
    Ours28.500.7848
    Table 1. PSNR and SSIM of different module combinations on Set14 dataset
    MethodPSNR /dB
    ResNeXt27.047
    ResNeXt+SE27.125
    ResNeXt+SimAM27.168
    Table 2. PSNR value of different module combinations on BSD100 dataset
    DatasetScaleBicubicSRCNNESPCNSRGANESRGANXLSROurs
    Set5428.4129.1529.6629.8230.4730.8731.62
    Set14426.0926.3226.8627.3326.6127.6928.50
    BSD100425.9526.3226.4526.6425.3227.0727.34
    Table 3. Average PSNR of different SR algorithms on three test sets at 4× magnification factor
    DatasetScaleBicubicSRCNNESPCNSRGANESRGANXLSROurs
    Set540.81280.82910.83360.84710.85180.87380.8880
    Set1440.71840.73610.73010.75170.71390.77290.7848
    BSD10040.67120.69210.67540.70110.65050.71920.7277
    Table 4. Average SSIM of different SR algorithms on three test sets at 4× magnification factor
    Yanfei Peng, Manting Zhang, Pingjia Zhang, Jian Li, Lirui Gu. Single-Image Super-Resolution Reconstruction Aggregating Residual Attention Network[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010017
    Download Citation