Low-Light Image Enhancement Based on Attention Mechanism and Convolutional Neural Networks

Fig. 4. Qualitative comparison of synthetic low-light images obtained by different algorithms. (a) Image “flowersonih35”; (b) image “plane”; (c) image “house”; (d) image “lighthouse”

Download full size

Fig. 5. Qualitative comparison of different algorithms on real low-light images. (a)(c) Images from DICM dataset; (b) image from LIME dataset; (d) image from MEF dataset

Download full size

Table 1. Performance comparison of different loss functions

View table

Table 1. Performance comparison of different loss functions

Image	PSNR /dB			SSIM
Image	L_MSE	L_SSIM	L_total	L_MSE	L_SSIM	L_total
img151	23.0820	23.4980	25.8330	0.7353	0.8126	0.8161
img165	22.7292	25.3378	26.5570	0.7832	0.8118	0.8115
img167	25.4330	22.8350	27.4850	0.8459	0.8929	0.9000
img168	26.3010	19.4620	26.1940	0.8943	0.9110	0.9253
img169	25.0620	24.9640	26.1580	0.8328	0.8734	0.8752
Average±SD	24.5200±1.3800	23.2200±2.0900	26.4500±0.5700	0.8180±0.0550	0.8600±0.0410	0.8660±0.0450

Table 2. Quantitative comparison of synthetic low-light images obtained by different algorithms

View table

Table 2. Quantitative comparison of synthetic low-light images obtained by different algorithms

Image	MSE	PSNR /dB	SSIM	TMQI	GMSD
Input image	7600.19	9.83	0.46	0.82	0.115
Image obtained by CLAHE	2799.89	15.75	0.76	0.83	0.087
Image obtained by NPE^[7]	463.44	22.06	0.85	0.87	0.034
Image obtained by LIME^[8]	1885.24	16.34	0.79	0.82	0.052
Image obtained by LLCNN^[14]	463.49	22.17	0.87	0.86	0.054
Image obtained by Ma^[15]	417.77	22.64	0.82	0.88	0.085
Image obtained by our method	146.56	26.72	0.88	0.88	0.033

Table 3. Quantitative comparison of different algorithms on real low-light images
View table
Table 3. Quantitative comparison of different algorithms on real low-light images
Method IE NIQE LOE SSEQ
CLAHE 7.59 8.08 1279.36 15.64
NPE^[7] 7.44 6.64 997.36 19.79
LIME^[8] 7.55 8.24 875.33 13.50
LLCNN^[14] 7.21 7.42 805.73 17.93
Ma^[15] 7.27 7.06 942.97 14.84
Ours 7.56 6.29 664.76 11.06

Tools

Get Citation

Copy Citation Text

Ruoyou Wu, Dexing Wang, Hongchun Yuan. Low-Light Image Enhancement Based on Attention Mechanism and Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2020, 57(20): 201022

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Mar. 2, 2020

Accepted: Apr. 15, 2020

Published Online: Oct. 14, 2020

The Author Email: Dexing Wang (dawang@shou.edu.cn), Hongchun Yuan (dawang@shou.edu.cn)

DOI:10.3788/LOP57.201022

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Performance comparison of different loss functions

Table 1. Performance comparison of different loss functions

Table 2. Quantitative comparison of synthetic low-light images obtained by different algorithms

Table 2. Quantitative comparison of synthetic low-light images obtained by different algorithms

Table 3. Quantitative comparison of different algorithms on real low-light images

Table 3. Quantitative comparison of different algorithms on real low-light images