Infrared Image Derivation Method for Generative Adversarial Network
243
4.2 Comparison Results
The experimental results are shown in Fig. 5. The pix2pix [7] method is a method
of infrared image derivation using a visible image alone. It is found that the visible
image is derived better in the first three channels, and the target derivative effect is more
realistic. Compared with GC, GD also lost the target details after using the previous
feature fusion, but the contour remains intact. and the distinction between target and
background is obvious.
Fig. 5. Comparison of Test Results
During the research, it is found that the image derived from GD and the temperature
vector used in ThermalGan. The derived average temperature infrared image is very
close, and ThermalGan treats the result of the real infrared image minus the average
temperature infrared image as a relative temperature infrared image.
The experimental results of GA, GB, GC, GD and the real infrared image were
calculated by Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM).
The average value is shown in Fig. 6.
Fig. 6. Objective Index
5 Conclusion
After designing and testing four generators, it is found that GB can use the light image as
the main channel of the image, and the target segmentation map as the image expansion
243
4.2 Comparison Results
The experimental results are shown in Fig. 5. The pix2pix [7] method is a method
of infrared image derivation using a visible image alone. It is found that the visible
image is derived better in the first three channels, and the target derivative effect is more
realistic. Compared with GC, GD also lost the target details after using the previous
feature fusion, but the contour remains intact. and the distinction between target and
background is obvious.
Fig. 5. Comparison of Test Results
During the research, it is found that the image derived from GD and the temperature
vector used in ThermalGan. The derived average temperature infrared image is very
close, and ThermalGan treats the result of the real infrared image minus the average
temperature infrared image as a relative temperature infrared image.
The experimental results of GA, GB, GC, GD and the real infrared image were
calculated by Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM).
The average value is shown in Fig. 6.
Fig. 6. Objective Index
5 Conclusion
After designing and testing four generators, it is found that GB can use the light image as
the main channel of the image, and the target segmentation map as the image expansion