Optics and Precision Engineering, Volume. 33, Issue 7, 1152(2025)
Infrared and visible image fusion based on multi-scale spatial attention complementary
Figures & Tables(17)
Fig. 1. Overall framework of progressive infrared and visible light image fusion network based on multi-level differential complementary multi-scale spatial attention
Fig. 2. Multi-scale spatial attention multi-level complementary module（MSAMCM）
Fig. 4. Comparison experimental results of sample 559 based on MSRS dataset for daytime scence （The red-framed area is a zoomed-in display of trees， and the green-framed area is a zoomed-in display of houses and windows in the fused image， color figures is showed in PDF file）
Fig. 5. Comparison experimental results of sample 808 based on MSRS dataset for nighttime scene （The red framed area is a zoomed-in view of pedestrians and notice boards， and the green framed area is a zoomed-in view of manhole covers in the fused image， color figures is showed in PDF file）
Fig. 6. Comparison of MI index for different image confusion methods under different cumulative propagation on MSRS dataset
Fig. 7. Comparison of VIF index for different image confusion methods under different cumulative propagation on MSRS dataset
Fig. 8. Comparison experimental results of sample 08835 on RoadScene dataset （The red-framed and green-framed areas are zoomed-in display of roads and bicycles， and pedestrians respectively in the fused image， color figures is showed in PDF file）
Fig. 9. Comparison experimental results of sample 08749 on RoadScene dataset （The red-framed and green-framed areas are zoomed-in views of road and rear door of car respectively in the fused image， color figures is showed in PDF file）
Fig. 10. Comparison experimental results of sample 05027 based on RoadScene dataset （The red-framed and green-framed areas are zoomed-in views of road and manholecover respectively in the fused image， color figures is showed in PDF file）
Fig. 11. Comparison experimental results of sample 06832 based on RoadScene dataset （The red-framed and green-framed areas are zoomed-in views of pedestrians and car as a whole in the fused image， color figures is showed in PDF file）
Fig. 12. Comparison of MI index for different confusion methods under different cumulative propagation based on RoadScene dataset
Fig. 13. Comparison of VIF index for different confusion methods under different cumulative propagation on RoadScene dataset
Fig. 14. Ablation results of fusion images in three typical scenes
Table 1. Kernel size， output channel， and activation function of all convolutional layers in feature extractor and image reconstructor
View tableView in ArticleTable 1. Kernel size， output channel， and activation function of all convolutional layers in feature extractor and image reconstructor
Layer Feature extractor Image reconstructor Kernel size Output channel Activation function Kernel size Output channel Activation function Conv 1 1×1 16 LeakyRelu 3×3 256 Leaky Relu Conv 2 3×3 16 LeakyRelu 3×3 128 Leaky Relu Conv 3 3×3 32 LeakyRelu 3×3 64 Leaky Relu Conv 4 3×3 64 LeakyRelu 3×3 32 Leaky Relu Conv 5 3×3 128 LeakyRelu 1×1 1 Tanh
Table 2. Comparison of mean and standard deviation of MI and VIF indicators between existing methods and proposed method on MSRS dataset
View tableView in ArticleTable 2. Comparison of mean and standard deviation of MI and VIF indicators between existing methods and proposed method on MSRS dataset
Method MI VIF Densefuse 2.392 8±0.486 2 0.887 9±0.176 0 FusionGAN 1.979 0±0.280 1 0.498 4±0.142 4 IFCNN 1.736 8±0.506 4 0.731 5±0.061 7 LRRNet 2.264 2±0.807 7 0.489 5±0.054 9 SDNet 1.915 9±0.331 6 0.577 2±0.059 7 PIAFusion 3.370 7±1.009 1 0.999 3±0.067 3 IVF-MSAC 3.515 2 ±1.013 6 1.051 4 ±0.082 2
Table 3. Comparison of average and standard deviation of MI and VIF indicators between existing methods and proposed method on RoadScene dataset
View tableView in ArticleTable 3. Comparison of average and standard deviation of MI and VIF indicators between existing methods and proposed method on RoadScene dataset
Method MI VIF Densefuse 2.522 9±0.419 7 0.479 7±0.040 8 FusionGAN 2.492 6±0.306 3 0.320 2±0.023 2 IFCNN 2.347 1±0.391 6 0.473 3±0.038 6 LRRNet 2.608 1±0.411 7 0.437 2±0.064 4 SDNet 2.953 5±0.432 3 0.493 0±0.033 5 PIAFusion 3.519 7±0.331 9 0.568 9±0.077 7 IVF-MSAC 3.678 1 ±0.378 5 0.582 3 ±0.085 5
Tools
Get Citation
Copy Citation Text
Yongxing ZHANG, Bowen LIAN, Naiting GU, Fangzhao LI, Yang LI. Infrared and visible image fusion based on multi-scale spatial attention complementary[J]. Optics and Precision Engineering, 2025, 33(7): 1152
Paper Information
Category:
Received: Dec. 31, 2024
Accepted: --
Published Online: Jun. 23, 2025
The Author Email: Naiting GU (gnt7328@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20