Infrared and Laser Engineering, Volume. 53, Issue 8, 20240198(2024)

Spatial and frequency domain feature decoupling for infrared and visible image fusion

Yan FAN1, Qiao LIU1, Di YUAN2, and Yunpeng LIU3
Author Affiliations
  • 1National Center for Applied Mathematics in Chongqing, Chongqing 401331, China
  • 2Guangzhou Institute of Technology, Xidian University, Guangzhou 710068, China
  • 3Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110169, China
  • show less
    Figures & Tables(15)
    (a) Framework of proposed algorithm: It contains a frequency decoupling branch, a spatial decoupling branch and a multi-frequency convolution attention block; (b) Framework of multi-frequency convolution attention block
    Comparison results of RoadScene dataset
    Comparison results of MSRS dataset
    Comparison results of TNO datase
    Visualization results of frequency domain disentangling branches
    Visualization results of spatial domain disentangling branches
    Visualization results of ablation comparison of different modules
    The visualization analysis and comparison of image pair "00959 N" in the MSRS dataset
    The visualization analysis and comparison of image pair "FLIR_04688" in the RoadScene dataset
    • Table 1. Objective evaluation results of RoadScene dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 1. Objective evaluation results of RoadScene dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      MethodEN↑SF↑AG↑MI↑SCD↑VIF↑$ \mathrm{\mathit{Q}}^{{{{A}{B}}}/\mathrm{\mathit{F}}} $SSIM↑
      RFN-Nest7.337.853.341.901.730.500.300.78
      SwinFusion7.0014.365.551.791.430.60.591.02
      SDNet7.3215.616.192.251.450.600.511.01
      U2Fusion6.8011.814.771.821.200.530.470.98
      Densefuse6.849.503.691.881.360.510.370.93
      FusionGAN7.078.733.371.911.130.370.260.59
      DATFusion6.7211.364.022.541.150.600.470.92
      Ours7.5516.886.332.081.770.620.490.96
    • Table 2. Objective evaluation results of MSRS dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 2. Objective evaluation results of MSRS dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      MethodEN↑SF↑AG↑MI↑SCD↑VIF↑$ {{Q}}^{{{{{A}{B}}}}/{{F}}} $SSIM↑
      RFN-Nest6.206.162.111.701.470.660.390.77
      SwinFusion6.0710.793.461.581.320.720.600.95
      SDNet5.258.672.671.180.990.500.380.72
      U2Fusion4.956.712.091.351.010.470.320.61
      Densefuse5.936.022.051.841.250.690.370.90
      FusionGAN5.434.351.451.310.980.440.140.50
      DATFusion6.4810.933.562.701.410.910.620.91
      Ours6.7112.283.942.651.690.940.640.95
    • Table 3. Objective evaluation results of TNO dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 3. Objective evaluation results of TNO dataset comparison experiment (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      MethodEN↑SF↑AG↑MI↑SCD↑VIF↑$ {{Q}}^{{A}{B}/{F}} $SSIM
      RFN-Nest6.965.872.661.461.780.560.330.81
      SwinFusion6.4910.224.001.311.640.620.551.04
      SDNet6.6911.644.561.561.560.580.430.97
      U2Fusion7.0011.865.001.391.780.620.430.94
      Densefuse6.828.993.541.591.780.660.451.02
      FusionGAN6.566.282.411.621.380.420.230.66
      DATFusion6.459.613.542.171.500.680.480.93
      Ours7.3113.145.191.841.790.750.500.99
    • Table 4. The impact of different spectral components on the fusion performance in MFCA (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 4. The impact of different spectral components on the fusion performance in MFCA (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      DatasetnEN$ \uparrow $SF$ \uparrow $AG$ \uparrow $MI$ \uparrow $SCD$ \uparrow $VIF$ \uparrow $$ {{Q}}^{{A}{B}/{F}}\uparrow $SSIM$ \uparrow $
      TNOn=26.9913.44.881.420.740.320.400.09
      n=47.0915.335.931.410.370.240.420.15
      n=87.1013.525.131.540.850.360.460.11
      n=167.3113.145.191.841.790.750.500.99
      n=327.2912.355.341.681.530.470.390.72
    • Table 5. Evaluation results of ablation experiments for each component. (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 5. Evaluation results of ablation experiments for each component. (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      $ \mathrm{D}\mathrm{a}\mathrm{t}\mathrm{a}\mathrm{s}\mathrm{e}\mathrm{t} $$ \mathrm{M}\mathrm{e}\mathrm{t}\mathrm{h}\mathrm{o}\mathrm{d}\mathrm{s} $$ \mathrm{E}\mathrm{N}\uparrow $$ \mathrm{A}\mathrm{G}\uparrow $$ \mathrm{S}\mathrm{S}\mathrm{I}\mathrm{M}\uparrow $$ \mathrm{V}\mathrm{I}\mathrm{F}\uparrow $$ {{Q}}^{{A}{B}/{F}}\uparrow $
      TNOSpa+MFCA7.064.640.980.720.50
      Fre+MFCA7.154.810.990.720.51
      Fre+Spa7.185.041.010.760.53
      Fre+Spa+MFCA7.315.191.020.780.55
      RoadSceneSpa+MFCA7.345.690.960.660.52
      Fre+MFCA7.364.940.970.690.50
      Fre+Spa7.506.330.990.670.50
      Fre+Spa+MFCA7.556.691.000.710.55
      MSRSSpa+MFCA6.673.740.970.970.65
      Fre+MFCA6.733.660.991.040.66
      Fre+Spa6.713.910.961.010.65
      Fre+Spa+MFCA6.803.921.001.040.70
    • Table 6. Evaluation results of the ablation study on loss functions (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      View table
      View in Article

      Table 6. Evaluation results of the ablation study on loss functions (Bold, underlined, and italicized correspond to the first, second, and third places respectively)

      $ \mathrm{D}\mathrm{a}\mathrm{t}\mathrm{a}\mathrm{s}\mathrm{e}\mathrm{t} $$ \mathrm{L}\mathrm{o}\mathrm{s}\mathrm{s} $$ \mathrm{E}\mathrm{N}\uparrow $$ \mathrm{M}\mathrm{I}\uparrow $$ \mathrm{S}\mathrm{S}\mathrm{I}\mathrm{M}\uparrow $$ \mathrm{V}\mathrm{I}\mathrm{F}\uparrow $$ {{Q}}^{{A}{B}/{F}}\uparrow $
      TNO$ {\mathcal{L}}_{{\mathrm{spa}}}+{\mathcal{L}}_{{\mathrm{recon}}} $7.191.360.150.370.39
      $ {\mathcal{L}}_{{\mathrm{fre}}}+{\mathcal{L}}_{{\mathrm{recon}}} $7.121.730.940.600.40
      $ {\mathcal{L}}_{{\mathrm{fre}}}+{\mathcal{L}}_{{\mathrm{spa}}}+{\mathcal{L}}_{{\mathrm{recon}}} $7.311.841.020.780.55
    Tools

    Get Citation

    Copy Citation Text

    Yan FAN, Qiao LIU, Di YUAN, Yunpeng LIU. Spatial and frequency domain feature decoupling for infrared and visible image fusion[J]. Infrared and Laser Engineering, 2024, 53(8): 20240198

    Download Citation

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

    Category:

    Received: May. 8, 2024

    Accepted: --

    Published Online: Oct. 29, 2024

    The Author Email:

    DOI:10.3788/IRLA20240198

    Topics