• Laser & Optoelectronics Progress
  • Vol. 58, Issue 14, 1410008 (2021)
Xueyan Zhang* and Yanwei Pang
Author Affiliations
  • School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP202158.1410008 Cite this Article Set citation alerts
    Xueyan Zhang, Yanwei Pang. Multiscale Nonlocal Neural Network for Single Image Deraining[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410008 Copy Citation Text show less
    Proposed network architecture
    Fig. 1. Proposed network architecture
    Initial module
    Fig. 2. Initial module
    Fusion module
    Fig. 3. Fusion module
    CCA block
    Fig. 4. CCA block
    Reconstruction module
    Fig. 5. Reconstruction module
    Derained results on Test1200 and Test1000 synthetic dataset. (a) Rain image; (b) DerainNet; (c) DID-MDN; (d) UMRL; (e) proposed method; (f) ground truth
    Fig. 6. Derained results on Test1200 and Test1000 synthetic dataset. (a) Rain image; (b) DerainNet; (c) DID-MDN; (d) UMRL; (e) proposed method; (f) ground truth
    De-rained results on real-world scenarios. (a) Rain image; (b) DerainNet; (c) DID-MDN; (d) UMRL; (e) proposed method
    Fig. 7. De-rained results on real-world scenarios. (a) Rain image; (b) DerainNet; (c) DID-MDN; (d) UMRL; (e) proposed method
    LayerParameter
    Conv-LSTMInput gateConv (3×3, 32)Stride is 1Sigmoid
    Forget gateConv (3×3, 32)Stride is 1Sigmoid
    Cell stateConv (3×3, 32)Stride is 1Sigmoid
    Output gateConv (3×3, 32)Stride is 1Sigmoid
    Table 1. Parameter settings of Conv-LSTM module
    ParameterModel 1Model 2Model 3Model 4Proposed method
    PSNR /dB28.7629.2826.1428.9530.16
    SSIM0.870.900.840.880.92
    Table 2. Quantitative comparation results of five models on Test1200
    DatasetRain imageDerainNetDDNDID-MDNUMRLProposed method
    PSNR /dB SSIMPSNR /dB SSIMPSNR /dB SSIMPSNR /dB SSIMPSNR /dB SSIMPSNR /dB SSIM
    Test120021.15 0.7722.07 0.8427.33 0.9027.95 0.9129.42 0.9130.16 0.92
    Test100019.31 0.7719.73 0.8325.63 0.8826.08 0.9026.47 0.9126.88 0.92
    Table 3. Quantitative evaluation results in terms of PSNR and SSIM
    Xueyan Zhang, Yanwei Pang. Multiscale Nonlocal Neural Network for Single Image Deraining[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410008
    Download Citation