Optics and Precision Engineering, Volume. 30, Issue 20, 2489(2022)

Multi-scale dense feature fusion network for image super-resolution

Deqiang CHENG1... Jiamin ZHAO1, Qiqi KOU2,*, Liangliang CHEN1 and Chenggong HAN1 |Show fewer author(s)
Author Affiliations
  • 1School of Information and Control Engineering, China University of Mining and Technology, Xuzhou226, China
  • 2School of Computer Science and Technology, China University of Mining and Technology, Xuzhou1116, China
  • show less
    Figures & Tables(14)
    Multi-scale dense feature fusion network
    Multi-scale feature fusion residual block
    Upsampling network
    Reconstruction results of Img040 in Urban100
    Reconstruction results of Img056 in Urban100
    Reconstruction results of Img092 in Urban100
    PSNR results of different D, C and G models
    Reconstruction results of Img056 in Urban100
    Reconstruction results of Img081 in Urban100
    PSNR and parameters on Set5 (×4) dataset of different models
    • Table 1. Effects of different connection modes and different feature extraction modules on model performance

      View table
      View in Article

      Table 1. Effects of different connection modes and different feature extraction modules on model performance

      Model层次特征融合密集特征融合PSNR/dBSSIM
      NSFB×32.300.929 9
      ×32.350.930 5
      Resblock×31.500.922 0
      ×31.530.922 2
      RDN×32.220.928 9
      ×32.290.929 6
      MSRB×32.290.929 7
      ×32.360.930 4
      MFRB×32.350.930 3
      ×32.410.932 2
    • Table 2. Index comparison under benchmark dataset when scaling factor is 2, 3 and 4

      View table
      View in Article

      Table 2. Index comparison under benchmark dataset when scaling factor is 2, 3 and 4

      模型

      缩放

      因子

      Set5Set14BSD100Urban100
      PSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIM
      Bicubic×233.660.929 930.240.868 729.560.843 126.880.840 1
      SRCNN×236.660.954 232.430.906 331.360.887 929.500.894 6
      VDSR×237.540.958 733.030.912 431.900.896 030.760.914 0
      DRCN×237.630.958 433.060.910 831.850.894 730.760.914 7
      LapSRN×237.530.959 133.080.910 931.800.894 930.410.911 2
      MSRN×238.080.960 533.740.917 032.230.901 332.220.932 6
      IMDN×238.000.960 533.630.917 732.190.899 632.170.928 3
      OISR-SK2×238.120.960 933.800.919 332.260.900 632.480.931 7
      LatticeNet×238.150.961 033.780.919 332.250.900 532.430.930 2
      DID-D5×238.150.961 033.770.919 032.270.900 632.380.930 5
      MDFN×238.140.961 033.830.919 632.270.900 632.410.931 0
      Bicubic×330.390.868 227.540.773 627.210.738 424.460.734 4
      SRCNN×332.750.909 029.300.821 528.410.786 326.240.798 9
      VDSR×333.660.921 329.770.831 428.820.797 627.140.827 9
      DRCN×333.850.921 529.890.831 728.810.795 427.160.831 1
      LapSRN×333.820.922 729.890.832 028.830.797 327.080.827 2
      MSRN×334.380.926 230.340.839 529.080.804 128.080.855 4
    • Table 2. Index comparison under benchmark dataset when scaling factor is 2, 3 and 4

      View table
      View in Article

      Table 2. Index comparison under benchmark dataset when scaling factor is 2, 3 and 4

      模型

      缩放

      因子

      Set5Set14BSD100Urban100
      PSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIM
      IMDN×334.360.927 030.320.841 729.090.804 628.170.851 9
      OISR-SK2×334.550.928 230.460.844 329.180.807 528.500.859 7
      LatticeNet×334.530.928 130.390.842 429.150.805 928.330.853 8
      DID-D5×334.550.928 030.490.844 629.190.806 928.390.856 6
      MDFN×334.600.928 430.500.844 929.210.807 528.520.859 1
      Bicubic×428.420.810 426.000.701 925.960.667 423.140.657 0
      SRCNN×430.480.862 827.490.750 326.900.710 124.530.722 1
      VDSR×431.350.883 028.010.768 027.290.725 125.180.754 3
      DRCN×431.560.881 028.150.762 027.240.715 025.150.753 0
      LapSRN×431.540.885 528.190.772 027.320.728 025.210.755 3
      MSRN×432.070.890 328.600.775 127.520.727 326.040.789 6
      IMDN×432.210.894 828.580.781 127.560.735 326.040.783 8
      OISR-SK2×432.320.896 528.720.784 327.660.739 026.370.795 3
      LatticeNet×432.300.896 228.680.783 027.620.736 726.250.787 3
      DID-D5×432.330.896 828.750.785 227.680.738 626.360.793 3
      MDFN×432.410.897 628.780.786 027.690.739 326.390.794 4
    • Table 3. Complexity and performance of different models

      View table
      View in Article

      Table 3. Complexity and performance of different models

      模型参数量/MFlops/GPSNR/dBSSIM
      MSRN6.08107.2732.070.890 3
      OISR-SK25.51117.4132.320.896 5
      DID-D55.2193.0332.330.896 8
      MDFN4.8987.7632.410.897 6
    Tools

    Get Citation

    Copy Citation Text

    Deqiang CHENG, Jiamin ZHAO, Qiqi KOU, Liangliang CHEN, Chenggong HAN. Multi-scale dense feature fusion network for image super-resolution[J]. Optics and Precision Engineering, 2022, 30(20): 2489

    Download Citation

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

    Category: Information Sciences

    Received: May. 9, 2022

    Accepted: --

    Published Online: Oct. 27, 2022

    The Author Email: KOU Qiqi (kouqiqi@cumt.edu.cn)

    DOI:10.37188/OPE.20223020.2489

    Topics