• Laser & Optoelectronics Progress
  • Vol. 58, Issue 16, 1610006 (2021)
Yue Wang, Dexing Wang*, Hongchun Yuan**, Ruoyou Wu, and Peng Gong
Author Affiliations
  • School of Information, Shanghai Ocean University, Shanghai 201306, China
  • show less
    DOI: 10.3788/LOP202158.1610006 Cite this Article Set citation alerts
    Yue Wang, Dexing Wang, Hongchun Yuan, Ruoyou Wu, Peng Gong. Underwater Image Enhancement Based on Pyramid Attention Mechanism and Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610006 Copy Citation Text show less
    Feature pyramid attention module
    Fig. 1. Feature pyramid attention module
    Structure of FPAGAN
    Fig. 2. Structure of FPAGAN
    Generative network structure
    Fig. 3. Generative network structure
    Discriminative network structure
    Fig. 4. Discriminative network structure
    Results of comparative experiments without FPA module and with FPA module. (a) Underwater images; (b) GAN; (c) FPAGAN
    Fig. 5. Results of comparative experiments without FPA module and with FPA module. (a) Underwater images; (b) GAN; (c) FPAGAN
    Qualitative comparison of different methods on test set A. (a) Underwater images; (b) GC; (c) UDCP; (d) LDCP; (e) UWCNN; (f) FUnIE-GAN; (g) proposed method; (h) reference
    Fig. 6. Qualitative comparison of different methods on test set A. (a) Underwater images; (b) GC; (c) UDCP; (d) LDCP; (e) UWCNN; (f) FUnIE-GAN; (g) proposed method; (h) reference
    Qualitative comparison of different methods on test set B. (a) Underwater images; (b) GC; (c) UDCP; (d) LDCP; (e) UWCNN; (f) FUnIE-GAN; (g) proposed method
    Fig. 7. Qualitative comparison of different methods on test set B. (a) Underwater images; (b) GC; (c) UDCP; (d) LDCP; (e) UWCNN; (f) FUnIE-GAN; (g) proposed method
    MethodPSNRSSIM
    GAN21.93280.7330
    FPAGAN22.39840.7418
    Table 1. Experimental results obtained by GAN and FPAGAN on test set A
    MethodUIQMIENIQE
    GAN2.90864.688942.5059
    FPAGAN2.94574.692537.6927
    Table 2. Experimental results obtained by GAN and FPAGAN on test set B
    MetricsGCUDCPLDCPUWCNNFUnIE-GANOurs
    PSNR15.183013.223113.979516.134419.357422.3984
    SSIM0.64950.53540.54220.60610.69230.7418
    Table 3. Quantitative comparison of different methods on test set A
    MetricsGCUDCPLDCPUWCNNFUnIE-GANOurs
    UIQM2.33611.67662.09342.22102.34182.9457
    IE4.22884.46784.54524.36994.65044.6925
    NIQE28.606731.021027.712139.856041.168337.6927
    OG-IQA-0.4489-0.6760-0.7238-0.5479-0.7475-0.8001
    Table 4. Quantitative comparison of different methods on test set B
    Yue Wang, Dexing Wang, Hongchun Yuan, Ruoyou Wu, Peng Gong. Underwater Image Enhancement Based on Pyramid Attention Mechanism and Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610006
    Download Citation