• Acta Optica Sinica
  • Vol. 39, Issue 10, 1010001 (2019)
Yong Chen*, Hongguang Guo, and Yapeng Ai
Author Affiliations
  • School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China
  • show less
    DOI: 10.3788/AOS201939.1010001 Cite this Article Set citation alerts
    Yong Chen, Hongguang Guo, Yapeng Ai. Single Image Dehazing Method Based on Multi-Scale Convolution Neural Network[J]. Acta Optica Sinica, 2019, 39(10): 1010001 Copy Citation Text show less
    Physical model of atmospheric scattering
    Fig. 1. Physical model of atmospheric scattering
    MSDN model diagram
    Fig. 2. MSDN model diagram
    Comparison of activation functions. (a) ReLU activation function; (b) PReLU activation function
    Fig. 3. Comparison of activation functions. (a) ReLU activation function; (b) PReLU activation function
    Algorithmic steps in this paper
    Fig. 4. Algorithmic steps in this paper
    Training data set. (a) Indoor data set ITS; (b) outdoor data set OTS
    Fig. 5. Training data set. (a) Indoor data set ITS; (b) outdoor data set OTS
    Experimental results of synthesizing hazy images. (a) Hazy image; (b) standard haze-free image; (c) method in Ref. [7]; (d) method in Ref. [11]; (e) method in Ref. [12]; (f) method in Ref. [13]; (g) method in Ref. [14]; (h) proposed method
    Fig. 6. Experimental results of synthesizing hazy images. (a) Hazy image; (b) standard haze-free image; (c) method in Ref. [7]; (d) method in Ref. [11]; (e) method in Ref. [12]; (f) method in Ref. [13]; (g) method in Ref. [14]; (h) proposed method
    Experimental results of real outdoor hazy images. (a) Hazy images; (b) method in Ref.[7]; (c) method in Ref.[11]; (d) method in Ref.[12]; (e) method in Ref.[13]; (f) method in Ref.[14]; (e) proposed method
    Fig. 7. Experimental results of real outdoor hazy images. (a) Hazy images; (b) method in Ref.[7]; (c) method in Ref.[11]; (d) method in Ref.[12]; (e) method in Ref.[13]; (f) method in Ref.[14]; (e) proposed method
    TypeConv
    Filter size3×35×57×7
    Filter number555
    Pad000
    Stride111
    Table 1. Parameter table of multi-scale feature extraction kernel
    ImageNo.Method in Ref.[7]Method in Ref.[11]Method in Ref.[12]Method in Ref.[13]Method in Ref.[14]Proposed method
    PSNR /dBSSIM /%PSNR /dBSSIM /%PSNR /dBSSIM /%PSNR /dBSSIM /%PSNR /dBSSIM /%PSNR /dBSSIM /%
    123.535785.2617.542459.5120.179672.3726.124785.3322.988879.4828.821886.41
    219.304480.0417.755272.3621.665584.0722.197788.7020.744287.3923.602791.66
    317.805179.6016.952172.8720.395981.2322.741489.3519.048684.0726.069189.73
    420.127782.7819.053479.3821.562384.4221.844486.4219.420281.6824.340291.60
    520.282581.8521.144482.8517.884879.2927.608193.5425.441090.6129.128594.78
    Table 2. Analysis of experimental data of synthetic hazy images
    ImageNo.Method in Ref.[7]Method in Ref.[11]Method in Ref.[12]Method in Ref.[13]Method in Ref.[14]Proposed method
    IEAGIEAGIEAGIEAGIEAGIEAG
    17.055514.647.065218.347.398418.527.244517.227.404820.267.682123.32
    27.515514.687.304917.047.819218.837.418613.937.665617.027.926618.99
    37.34278.487.473710.897.877111.787.70438.627.61209.217.893512.08
    47.56889.187.425011.937.851513.237.76089.627.778610.577.981514.31
    57.253814.917.721318.177.341017.937.174610.437.377613.627.869019.41
    67.16678.927.837110.407.716810.187.02637.857.28628.767.893710.47
    77.262510.947.113612.967.683413.717.32008.677.51899.597.727414.06
    86.27215.887.14597.327.35617.556.83635.976.98186.397.45347.57
    Table 3. Analysis of experimental data of outdoor hazy images
    MethodExperiment
    IndoorOutdoot
    Method in Ref.[7]6.876.89
    Method in Ref.[11]3.413.65
    Method in Ref.[12]1.962.08
    Method in Ref.[13]1.211.26
    Method in Ref.[14]1.581.92
    Proposed method1.091.18
    Table 4. Running time of different algorithms for experimental imagess
    Yong Chen, Hongguang Guo, Yapeng Ai. Single Image Dehazing Method Based on Multi-Scale Convolution Neural Network[J]. Acta Optica Sinica, 2019, 39(10): 1010001
    Download Citation