Fig 1.
Retinex schematic diagram.
Fig 2.
Overall network structure.
Fig 3.
Structure of the decomposition network.
Fig 4.
Results of adding SSIM loss function to KinD.
Fig 5.
Comparison of ReLU function and GELU function in decomposition network.
Fig 6.
Structure of the reflective component denoising network.
Fig 7.
Results of adding CA attention to Unet3+.
Fig 8.
Structure of the illumination component enhancement network.
Fig 9.
Comparison results of different iteration times n.
Table 1.
Comparison results of different iteration times n.
Fig 10.
Comparison results of different methods on LOL test set.
Fig 11.
Comparison results of different methods on VE-LOL-L test set.
Fig 12.
Comparison results of different methods on DICM test set.
Fig 13.
Comparison results of different methods on MEF test set.
Fig 14.
Comparison results of different methods on SID test set.
Fig 15.
Comparison results of different methods on ELD test set.
Table 2.
Comparison results of different methods on LOL dataset.
Table 3.
Comparison results of different methods on VE-LOL-L dataset.
Table 4.
Comparison results of different methods on SID dataset.
Table 5.
Comparison results of different methods on ELD dataset.
Table 6.
Comparison results of different methods on DICM dataset.
Table 7.
Comparison results of different methods on MEF dataset.
Fig 16.
Ablation studies on the framework.
‘w/o Denois.’ denotes our method without reflective component denoising network. ‘w/o Enhance.’ denotes our method without illumination component enhancement network.
Table 8.
Comparison of ablation results for different modules.