Laser & Optoelectronics Progress, Volume. 62, Issue 2, 0237011(2025)
ULCF-Net: An Underwater Low-Illumination Image Enhancement Algorithm Based on a Cross-Scale Structure and Color Fusion
Figures & Tables(14)
Fig. 3. Structure of artifact suppression sampling module. (a) Process of coding end sampling; (b) process of inverse operation
Fig. 4. Cross-scale connections. (a) Original cross-scale connection; (b) cross-scale connection Ⅰ; (c) cross-scale connection Ⅱ
Fig. 7. Subjective comparison on test set A between proposed algoritm and low light image enhancement algorithms
Fig. 8. Subjective comparison on test set A between proposed algoritm and underwater image enhancement algorithms
Table 1. Comparison results on test set A between proposed algoritm and low light image enhancement algorithms
View tableTable 1. Comparison results on test set A between proposed algoritm and low light image enhancement algorithms
Algorithm PSNR /dB SSIM MSE LPIPS UIQM UCIQE LAU-Net［ 24 ］21.4683 0.8601 0.0078 0.1239 3.1583 0.5993 HWMNet［ 18 ］20.2836 0.8374 0.0105 0.1544 3.1796 0.5912 Retinexformer［ 25 ］20.7749 0.8565 0.0099 0.1246 3.0783 0.5895 Enlighten Anything［ 26 ］15.3281 0.7225 0.0324 0.2683 2.8506 0.5384 FourLLIE［ 19 ］21.6478 0.8537 0.0085 0.1273 3.2599 0.6063 CDAN［ 27 ］20.9618 0.8676 0.0094 0.1184 3.2080 0.6015 ULCF-Net （proposed） 23.2729 0.8781 0.0053 0.0923 3.1150 0.6106
Table 2. Comparison results on test set A between proposed algoritm and underwater image enhancement algorithms
View tableTable 2. Comparison results on test set A between proposed algoritm and underwater image enhancement algorithms
Algorithm PSNR /dB SSIM MSE LPIPS UIQM UCIQE LDS-Net［ 21 ］18.1289 0.7950 0.0167 0.1885 3.2921 0.5702 U-shape［ 28 ］22.5064 0.7736 0.0066 0.1635 3.2169 0.5807 Five A+［ 14 ］21.8892 0.8562 0.0078 0.1294 3.1693 0.5931 LANet［ 29 ］21.3721 0.8316 0.0092 0.1029 3.0498 0.6099 STSC［ 30 ］18.5456 0.5483 0.0164 0.1205 3.0935 0.5995 Shallow-UWnet［ 31 ］12.8915 0.6448 0.0608 0.2186 2.8725 0.2946 ULCF-Net （proposed） 23.2729 0.8781 0.0053 0.0923 3.1150 0.6106
Table 3. Objectively evaluate results on test set C by different algorithms
View tableTable 3. Objectively evaluate results on test set C by different algorithms
Algorithm UIQM UCIQE LDS-Net［ 21 ］2.7827 0.5672 HWMNet［ 18 ］2.4927 0.5898 Enlightn Anything［ 26 ］1.5338 0.5591 FourLLIE［ 19 ］2.7137 0.5842 CDAN［ 27 ］2.4573 0.5758 LAU-Net［ 24 ］2.2548 0.5830 STSC［ 30 ］2.5992 0.5861 Shallow-UWnet［ 31 ］1.8127 0.5335 ULCF-Net （proposed） 2.3703 0.5916
Table 4. Module to module ablation experiment
View tableTable 4. Module to module ablation experiment
BEM form CEM type PSNR /dB SSIM MSE BEM（a） BEM（b） SKFF DSMCEM 15.4485 0.6912 0.0305 √ 17.1021 0.7573 0.0216 √ √ 19.9057 0.8341 0.0116 √ √ 19.7985 0.8328 0.0119 √ 16.8362 0.7656 0.0248 √ √ 22.8939 0.8712 0.0059 √ √ 23.2729 0.8781 0.0053
Table 5. Cross-scale connection ablation experiment
View tableTable 5. Cross-scale connection ablation experiment
Cross-scale connection form PSNR /dB SSIM MSE Origin Ⅰ Ⅱ √ 22.8718 0.8699 0.0063 √ 23.1546 0.8701 0.0062 √ 23.2729 0.8781 0.0053
Tools
Get Citation
Copy Citation Text
Yang Tao, Hao Tan, Liqun Zhou. ULCF-Net: An Underwater Low-Illumination Image Enhancement Algorithm Based on a Cross-Scale Structure and Color Fusion[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0237011
Paper Information
Category: Digital Image Processing
Received: Apr. 30, 2024
Accepted: Jun. 3, 2024
Published Online: Dec. 17, 2024
The Author Email:
CSTR:32186.14.LOP241207
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20