• Laser & Optoelectronics Progress
  • Vol. 59, Issue 4, 0420002 (2022)
Yuanxue Xin2, Fengting Zhu2, Pengfei Shi1、2、*, Xin Yang2, and Runkang Zhou2
Author Affiliations
  • 1Jiangsu Key Laboratory of Power Transmission & Distribution Equipment Technology, Hohai University, Changzhou , Jiangsu 213022, China
  • 2College of Internet of Things Engineering, Hohai University, Changzhou , Jiangsu 213022, China
  • show less
    DOI: 10.3788/LOP202259.0420002 Cite this Article Set citation alerts
    Yuanxue Xin, Fengting Zhu, Pengfei Shi, Xin Yang, Runkang Zhou. Super-Resolution Reconstruction Algorithm of Images Based on Improved Enhanced Super-Resolution Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2022, 59(4): 0420002 Copy Citation Text show less
    Improved structures of ESRGAN feature extraction module. (a) RB; (b) RRDB; (c) DB
    Fig. 1. Improved structures of ESRGAN feature extraction module. (a) RB; (b) RRDB; (c) DB
    Basic block of generative network. (a) MDB with attention mechanism (MADB); (b) multi-scale feature extraction block(MFEB)
    Fig. 2. Basic block of generative network. (a) MDB with attention mechanism (MADB); (b) multi-scale feature extraction block(MFEB)
    Structure of the ECA module
    Fig. 3. Structure of the ECA module
    RR-MADB structure
    Fig. 4. RR-MADB structure
    Generative network structure
    Fig. 5. Generative network structure
    Discriminant network structure
    Fig. 6. Discriminant network structure
    2× super-resolution reconstruction results of each algorithm. (a) Original images; (b) high resolution images; (c) Bicubic;(d) FSRCNN; (e) ESRGAN; (f) Ours-L1; (g) Ours
    Fig. 7. 2× super-resolution reconstruction results of each algorithm. (a) Original images; (b) high resolution images; (c) Bicubic;(d) FSRCNN; (e) ESRGAN; (f) Ours-L1; (g) Ours
    4× super-resolution reconstruction results of each algorithm. (a) Original images; (b) high resolution images; (c) Bicubic;(d) FSRCNN; (e) ESRGAN; (f) Ours-L1; (g) Ours
    Fig. 8. 4× super-resolution reconstruction results of each algorithm. (a) Original images; (b) high resolution images; (c) Bicubic;(d) FSRCNN; (e) ESRGAN; (f) Ours-L1; (g) Ours
    DatasetScaleBicubicFSRCNNESRGANOurs-L1Ours
    Set531.78733.40432.63735.75833.193
    26.69027.56927.92830.34728.629
    Set1428.29829.48728.78731.38729.481
    24.23524.83324.51326.61424.786
    BSD10026.72526.76325.80126.96425.953
    23.70123.73022.47823.87222.836
    Table 1. Comparison of PSNR values of the 2× and 4× reconstructed results of each algorithm
    DatasetScaleBicubicFSRCNNESRGANOurs-L1Ours
    Set50.9080.9220.9040.9430.906
    0.7690.7850.7990.8630.818
    Set140.8090.8650.8370.8920.845
    0.6600.6830.6600.7450.670
    BSD1000.7360.7550.7250.7590.730
    0.5720.5800.5250.5900.541
    Table 2. Comparison of SSIM values of the 2× and 4× reconstructed results of each algorithm
    Yuanxue Xin, Fengting Zhu, Pengfei Shi, Xin Yang, Runkang Zhou. Super-Resolution Reconstruction Algorithm of Images Based on Improved Enhanced Super-Resolution Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2022, 59(4): 0420002
    Download Citation