Optics and Precision Engineering, Volume. 33, Issue 6, 916(2025)
Dual scale fusion image dehazing algorithm based on frequency-domain feature distillation
Figures & Tables(18)
Fig. 6. Visual comparison of different algorithms on SOTS indoor composite images
Fig. 7. Visual comparison of different algorithms on SOTS outdoor composite images
Fig. 8. Visual comparison of different algorithms on HSTS composite images
Fig. 9. Visual comparison of different algorithms on real-world images of I-Haze
Fig. 10. Visual comparison of different algorithms on the HSTS real image dataset
Fig. 11. Visual comparison of different algorithms on the Fattal dataset
Table 1. Comparison of model performance under different batch size
View tableView in ArticleTable 1. Comparison of model performance under different batch size
批次 PSNR SSIM 1 27.54 0.959 2 2 32.95 0.979 5 3 32.50 0.975 4
Table 2. Objective evaluation indicators of different algorithms on the SOTS dataset
View tableView in ArticleTable 2. Objective evaluation indicators of different algorithms on the SOTS dataset
算法 Indoor Outdoor PSNR SSIM PSNR SSIM DCP 16.35 0.783 1 17.27 0.857 0 AOD-Net 19.69 0.839 5 19.63 0.891 4 USID-Net 17.45 0.769 9 23.86 0.888 6 EPDN 25.17 0.925 4 20.28 0.838 3 DehazeFlow 25.88 0.839 4 27.84 0.960 0 MSTN 35.25 0.985 1 29.96 0.934 9 Ours 36.74 0.991 3 32.95 0.979 5
Table 3. Comparison of objective evaluation indicators of different algorithms on the I-HAZE dataset
View tableView in ArticleTable 3. Comparison of objective evaluation indicators of different algorithms on the I-HAZE dataset
算法 PSNR SSIM DCP 15.259 0.697 3 AOD-Net 14.732 0.673 3 USID-Net 15.165 0.620 9 EPDN 14.947 0.686 9 DehazeFlow 10.145 0.335 9 MSTN 16.173 0.755 2 Ours 16.273 0.749 2
Table 4. Comparison of objective evaluation indicators without reference images
View tableView in ArticleTable 4. Comparison of objective evaluation indicators without reference images
算法 IE ↑ NIQE ↓ Hazy 6.903 8 3.601 6 DCP 6.147 2 3.376 3 AOD-Net 7.170 0 3.273 6 USID-Net 7.387 4 3.154 7 EPDN 7.350 0 3.446 4 DehazeFlow 7.077 3 3.402 5 MSTN 7.172 4 3.092 1 Ours 7.369 0 3.2287
Table 5. Objective evaluation results of different models
View tableView in ArticleTable 5. Objective evaluation results of different models
模型 PSNR SSIM Base 27.78 0.9554 模型1 28.20 0.9551 模型2 30.01 0.9660 模型3 31.88 0.9746 模型4 32.87 0.9779 模型5 32.56 0.9771 Ours 32.95 0.9795
Table 6. Runtime of different algorithms
View tableView in ArticleTable 6. Runtime of different algorithms
算法 Fattal HSTS DCP 0.1289 0.3621 AOD-Net 0.5707 0.5725 USID-Net 0.7882 1.5497 EPDN 1.3090 1.8047 DehazeFlow 6.1540 80.7334 MSTN 0.5242 0.8147 Ours 1.1541 1.2482
Tools
Get Citation
Copy Citation Text
Qingjiang CHEN, Shuang YANG. Dual scale fusion image dehazing algorithm based on frequency-domain feature distillation[J]. Optics and Precision Engineering, 2025, 33(6): 916
Paper Information
Category:
Received: May. 31, 2024
Accepted: --
Published Online: Jun. 16, 2025
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20