Fig 1.
Schematic diagram of UMFFA structure.
Fig 2.
Composition of the RDAB and Basic Block.
Fig 3.
Multi-path channel attention and pixel attention modules.
Fig 4.
Input and pixel channel attention weight.
(a) Input and (b) Pixel channel weight.
Fig 5.
Multi-path channel attention weight.
Fig 6.
(a) Ground truth and (b) Hazy image.
Table 1.
Training platform and related parameters.
Table 2.
Comparison of the performance of different methods.
Table 3.
Ablation test setup.
Fig 7.
Changes in performance metrics of different methods in ablation tests.
(a) PSNR and (b) SSIM.
Table 4.
Comparison of ablation test performance metrics on the Highway data set (average on last 20 results).
Fig 8.
Effect of different β on the performance metrics PSNR/SSIM.
(a) PSNR and (b) SSIM.
Fig 9.
The effect of β on the performance metrics L∞ error.
Table 5.
Influence of β on performance metrics.
Table 6.
Comparison on Highway dataset with multi-channel pooling.
Table 7.
Comparison results with multi-channel pooling on the I-Haze dataset.
Fig 10.
Image 109 from dataset I-Haze.
Fig 11.
Image 132 from dataset I-Haze.
Fig 12.
Image 21 from dataset Highway.
Fig 13.
Image 28 from dataset Highway.