Unsupervised Dehazing Algorithm Based on Multi-Scale Features

Xiangsheng Sun; Guozhong Wang

doi:10.3788/LOP202158.1610024

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 16 >Page 1610024 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 16, 1610024 (2021)

Unsupervised Dehazing Algorithm Based on Multi-Scale Features

Xiangsheng Sun and Guozhong Wang^*

Author Affiliations

School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

show less

DOI: 10.3788/LOP202158.1610024 Cite this Article Set citation alerts

Xiangsheng Sun, Guozhong Wang. Unsupervised Dehazing Algorithm Based on Multi-Scale Features[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610024 Copy Citation Text

show less

Fig. 1. Overall structure of MFC-dehaze removal network

Download full size

Fig. 2. Network architecture of multiscale generator

Download full size

Fig. 3. Network structure of discriminator

Download full size

Fig. 4. Dehazing results of different algorithms in NYU-Depth dataset. (a) Original images; (b) Ref. [4]; (c) Ref. [7]; (d) Ref. [16]; (e) Ref. [17]; (f) proposed algorithm; (g) ground truth images

Download full size

Fig. 5. Dehazing results of different algorithms in I-HAZE and O-HAZE datasets. (a) Original images; (b) Ref. [4]; (c) Ref. [8]; (d) Ref. [15]; (e) Ref. [11]; (f) Ref. [14]; (g) proposed algorithm; (h) ground truth images

Download full size

Fig. 6. Real data comparison results of different algorithms. (a) Original images; (b) Ref. [4]; (c) Ref. [8]; (d) Ref. [15]; (e) Ref. [11]; (f) proposed algorithm

Download full size

Index	Ref. [4]	Ref. [7]	Ref. [8]	Ref. [14]	Ref. [15]	Ref. [16]	Ref. [17]	Proposed algorithm
PSNR /dB	16.84	18.98	19.28	21.65	18.63	19.41	24.89	27.83
SSIM	0.768	0.783	0.831	0.824	0.796	0.794	0.865	0.908

Table 1. Comparison results of different algorithms in NYU-Depth dataset

Index	Ref. [4]	Ref. [7]	Ref. [8]	Ref. [14]	Ref. [15]	Ref. [16]	Ref. [17]	Proposed algorithm
PSNR /dB	15.84	18.43	20.42	21.65	19.88	20.67	24.87	27.61
SSIM	0.761	0.766	0.811	0.807	0.853	0.864	0.853	0.921

Table 2. Comparison of results of different algorithms in I-HAZE dataset

Index	Ref. [4]	Ref. [7]	Ref. [8]	Ref. [14]	Ref. [15]	Ref. [16]	Ref. [17]	Proposed algorithm
PSNR /dB	14.61	19.28	20.14	22.72	22.68	21.35	25.14	26.93
SSIM	0.781	0.804	0.842	0.821	0.845	0.826	0.833	0.937

Table 3. Comparison of results of different algorithms in O-HAZE dataset

Index	cycleGAN	cycleGAN+spectral normalization	cycleGAN+four discriminator	Ours
PSNR /dB	22.71	23.63	25.88	26.92
SSIM	0.772	0.812	0.866	0.936

Table 4. Results of ablation experiments for network structure

Index	λ₁=5, λ₂=5.0, λ₃=5	λ₁=10, λ₂=0.5, λ₃=5	λ₁=10, λ₂=0.1, λ₃=5
PSNR /dB	21.32	23.78	26.92
SSIM	0.791	0.892	0.936

Table 5. Results of ablation experiments with different parameters

Xiangsheng Sun, Guozhong Wang. Unsupervised Dehazing Algorithm Based on Multi-Scale Features[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610024

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information