Optics and Precision Engineering, Volume. 32, Issue 10, 1538(2024)
Cross-level feature aggregation image enhancement with dual-path hybrid attention
Figures & Tables(11)
Fig. 5. Visual comparison of different algorithms on the LOL dataset
Fig. 6. Visual comparison of different algorithms on the MIT-Adobe 5K dataset
Fig. 7. Ablation experiment visualization of different network modules
Table 1. Objective evaluation results of different algorithms on the LOL dataset
View tableView in ArticleTable 1. Objective evaluation results of different algorithms on the LOL dataset
方法 PSNR/dB↑ SSIM↑ LPIPS↓ NIQE↓ BPDHE 10.936 0.415 0.439 7.985 CLAHE 13.414 0.551 0.361 6.636 LIME 16.758 0.564 0.395 8.058 BIMEF 13.875 0.575 0.326 7.699 RetinexNet 16.774 0.425 0.474 8.879 KinD 17.647 0.771 0.514 5.189 URetinex Net 19.841 0.826 0.235 4.722 DSLR 14.982 0.596 0.376 4.416 Uformer 18.547 0.721 0.321 4.443 ALL-E 18.216 0.763 0.219 4.200 EnlightenGAN 17.482 0.651 0.322 5.807 ZeroDCE 14.860 0.562 0.335 7.767 ZeroDCE++ 16.570 0.591 0.328 5.311 RUAS 18.226 0.717 0.270 6.340 SCI 14.780 0.522 0.339 6.617 Ours 22.347 0.850 0.178 4.153
Table 2. Objective evaluation results of different algorithms on the MIT-Adobe 5K dataset
View tableView in ArticleTable 2. Objective evaluation results of different algorithms on the MIT-Adobe 5K dataset
方法 PSNR/dB↑ SSIM↑ LPIPS↓ NIQE↓ BPDHE 12.855 0.572 0.441 5.280 CLAHE 15.955 0.608 0.375 5.353 LIME 13.303 0.749 0.319 4.172 BIMEF 17.968 0.797 0.298 4.598 RetinexNet 12.514 0.670 0.365 4.841 KinD 16.203 0.784 0.254 4.242 URetinex Net 14.184 0.754 0.242 3.867 DSLR 20.243 0.828 0.153 4.352 UFormer 21.917 0.870 0.185 3.961 DDNet 18.308 0.776 0.220 3.830 EnlightenGAN 17.905 0.836 0.238 3.865 ZeroDCE 15.931 0.766 0.219 3.882 ZeroDCE++ 14.611 0.405 0.231 3.850 RUAS 15.995 0.786 0.141 4.048 SCI 17.477 0.840 0.216 4.122 Ours 22.703 0.903 0.137 3.822
Table 3. Comparison results of evaluation indicators of different network modules
View tableView in ArticleTable 3. Comparison results of evaluation indicators of different network modules
模块 PSNR/dB↑ SSIM↑ Time/s↓ 无MDAR 21.505 0.861 0.100 无PMFB 21.862 0.878 0.108 无DHAB 22.512 0.896 0.109 无CFAM 21.173 0.732 0.106 无混合空洞卷积 21.901 0.890 0.110 Ours 22.703 0.903 0.112
Table 4. Average enhancement time of different method
View tableView in ArticleTable 4. Average enhancement time of different method
BPDHE CLAHE LIME BIMEF Retinex Net KinD URetinex Net DSLR 0.1041 0.1156 0.4914 0.1280 0.1200 0.0500 0.400 0.125 Running time↓ UFormer ALL-E Enlighten GAN Zero DCE Zero DCE++ RUAS SCI Ours 0.121 1.055 0.057 0.003 0.0019 0.006 0.001 0.112
Tools
Get Citation
Copy Citation Text
Heng YUAN, Xiaoxue WANG, Tinghao YAN, Shengchong ZHANG. Cross-level feature aggregation image enhancement with dual-path hybrid attention[J]. Optics and Precision Engineering, 2024, 32(10): 1538
Paper Information
Category:
Received: Nov. 21, 2023
Accepted: --
Published Online: Jul. 8, 2024
The Author Email: WANG Xiaoxue (wxx2585354184@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20