Optics and Precision Engineering, Volume. 31, Issue 17, 2584(2023)

Multidimensional attention mechanism and selective feature fusion for image super-resolution reconstruction

Jian WEN1... Jianfei SHAO1,*, Jie LIU2, Jianlong SHAO1, Yuhang FENG1 and Rong YE1 |Show fewer author(s)
Author Affiliations
  • 1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming650500, China
  • 2Yunnan Police Unmanned System Innovation Research Institute, Yunnan Police Officer Academy, Kunming6503, China
  • show less
    Figures & Tables(13)
    Schematic diagram of network structure integrating multi-dimensional attention mechanism and selective feature fusion
    Schematic diagram of selective kernel feature fusion module
    Schematic diagram of internal principle junction of heterogeneous residual module
    Comparison of PSNR ablation on Urban100 dataset magnified 2 times
    Comparison of Img015 magnified 2 times in Urban100 dataset
    Comparison of Img030 magnified 3 times in Urban100 dataset
    Comparison of Img039 magnified 4 times in Urban100 dataset
    PSNR and parameters on Set5(×4)dataset of different models
    • Table 1. Comparison results of different methods on Urban100 data set magnified by 2 times

      View table
      View in Article

      Table 1. Comparison results of different methods on Urban100 data set magnified by 2 times

      方法/评价指标对称组卷积模块互补卷积模块

      选择性内核

      特征融合模块

      特征增强

      残差模块

      PSNR/dBSSIM
      Plan A××30.430.921 6
      Plan B××31.260.915 7
      Plan C××31.150.911 7
      Plan D×31.790.924 5
      Plan E32.120.928 8
    • Table 2. Comparison results of different methods on 4 Baseline datasets enlarged by 2 times

      View table
      View in Article

      Table 2. Comparison results of different methods on 4 Baseline datasets enlarged by 2 times

      算 法Set5Set14BSDS100Urban100
      PSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIM
      Bicubic3233.660.929 930.240.868 829.560.84126.880.840 3
      A+3336.540.954 432.280.905 631.210.886 329.200.893 8
      RFL3436.540.953 732.260.904 031.160.884 029.110.890 4
      DASR3537.220.951 532.930.902 931.760.890 130.630.907 9
      KXNet3737.590.955 233.360.909 132.030.894 131.480.919 2
      RED303837.660.959 932.940.914 431.990.897 430.910.915 9
      DnCNN3637.580.959 033.030.912 831.900.896 130.740.913 9
      MemNet3937.780.959 733.280.914 232.080.897 831.310.919 5
      CARN1937.760.959 033.520.916 631.920.897 831.510.931 2
      CARN-M1937.530.958 333.260.914 132.090.896 030.830.923 3
      EEDS+1537.780.960 933.210.915 131.950.896 331.920.928 9
      MSDEPC4137.390.957 632.940.911 131.640.896 131.830.928 3
      CFSRCNN4237.790.959 133.510.916 532.110.898 832.070.819 0
      LESRCNN2237.650.958 633.320.914 831.950.896 431.450.920 6
      LESRCNN-S2237.570.958 233.300.914 531.950.896 531.450.920 7
      ACNet4337.720.958 833.410.916 032.060.897 831.790.924 5
      ACNet-B4337.600.958 433.320.915 131.970.897 031.570.922 2
      Ours37.830.959 433.530.917 232.140.898 732.120.930 6
    • Table 3. Comparison results of different methods on 4 Baseline datasets enlarged by 3 times

      View table
      View in Article

      Table 3. Comparison results of different methods on 4 Baseline datasets enlarged by 3 times

      算 法Set5Set14BSDS100Urban100
      PSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIM
      Bicubic3230.390.868 227.550.774 227.210.738 524.460.734 9
      A+3332.580.908 829.130.818 828.290.783 526.030.797 3
      RFL3432.430.905 729.050.816 428.220.780 625.860.790 0
      DASR3533.780.920 029.940.826 628.850.793 227.280.830 7
      KXNet3734.220.923 830.270.834 029.060.801 028.000.845 7
      RED303833.820.929 929.610.834 128.930.799 427.310.830 3
      DnCNN3633.750.929 029.810.832 128.850.798 127.150.827 6
      MemNet3934.090.924 830.000.835 028.960.800 127.560.837 6
      CARN1933.990.923 630.080.836 728.910.800 026.860.826 3
      CARN-M1934.290.925 530.290.840 729.060.803 427.380.840 4
      EEDS+1533.810.925 229.850.833 928.880.805 427.430.835 9
      MSDEPC4133.370.918 429.620.827 929.580.791 827.230.819 0
      CFSRCNN4233.240.925 630.270.841 029.030.803 528.040.849 6
      LESRCNN2234.930.923 130.120.838 028.910.800 527.700.841 5
      LESRCNN-S2233.050.923 830.160.838 428.940.801 227.760.842 4
      ACNet4334.140.924 730.190.839 828.980.802 327.970.848 2
      ACNet-B4334.070.924 330.150.838 628.970.801 627.880.844 7
      Ours34.350.926 830.360.841 729.110.804 728.190.853 1
    • Table 4. Comparison results of different methods on 4 Baseline datasets enlarged by 4 times

      View table
      View in Article

      Table 4. Comparison results of different methods on 4 Baseline datasets enlarged by 4 times

      算 法Set5Set14BSDS100Urban100
      PSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIMPSNR/dBSSIM
      Bicubic3228.420.810 426.000.702 725.960.667 523.140.657 7
      A+3330.280.860 327.320.749 126.820.708 724.320.718 3
      RFL3430.140.854 827.240.745 126.820.705 424.190.709 6
      DASR3531.680.885 428.260.766 827.400.728 625.490.762 1
      KXNet3731.940.891 228.610.778 227.590.733 026.180.787 3
      RED303831.510.886 927.860.771 827.400.729 025.350.758 7
      DnCNN3631.400.884 528.040.767 227.290.725 325.200.752 1
      MemNet3931.740.889 328.260.772 327.400.728 125.500.763 0
      CARN1931.920.890 328.420.776 227.440.730 425.630.768 8
      CARN-M1932.130.893 728.600.780 627.580.734 926.070.783 7
      EEDS+1531.530.886 928.130.769 827.350.726 325.450.762 9
      MSDEPC4131.050.879 727.790.758 127.100.719 325.230.758 6
      CFSRCNN4232.060.892 928.570.780 027.530.733 326.030.782 4
      LESRCNN2231.880.890 328.440.777 227.450.731 325.770.773 2
      LESRCNN-S2231.880.890 728.430.777 627.470.732 125.780.773 9
      ACNet4331.830.890 328.460.778 827.480.732 625.930.779 8
      ACNet-B4331.820.890 128.410.777 327.460.731 625.860.776 0
      Ours32.110.894 328.630.781 227.650.735 226.070.790 1
    • Table 5. Inference time of different models on Set5(×4)dataset

      View table
      View in Article

      Table 5. Inference time of different models on Set5(×4)dataset

      算 法推理时间/s
      CARN0.032 0
      CFSRCNN0.029 8
      MemNet14.690 0
      ACNet0.031 5
      DASR0.072 1
      KXNet0.642 9
      Ours0.051 6
    Tools

    Get Citation

    Copy Citation Text

    Jian WEN, Jianfei SHAO, Jie LIU, Jianlong SHAO, Yuhang FENG, Rong YE. Multidimensional attention mechanism and selective feature fusion for image super-resolution reconstruction[J]. Optics and Precision Engineering, 2023, 31(17): 2584

    Download Citation

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

    Category: Information Sciences

    Received: Nov. 22, 2022

    Accepted: --

    Published Online: Oct. 9, 2023

    The Author Email: SHAO Jianfei (469365367@qq.com)

    DOI:10.37188/OPE.20233117.2584

    Topics