• Laser & Optoelectronics Progress
  • Vol. 55, Issue 3, 031004 (2018)
Ming Zhang, Xiaoqi Lü*, Liang Wu, and Dahua Yu
Author Affiliations
  • School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou, Inner Mongolia 014010, China
  • show less
    DOI: 10.3788/LOP55.031004 Cite this Article Set citation alerts
    Ming Zhang, Xiaoqi Lü, Liang Wu, Dahua Yu. Multiplicative Denoising Method Based on Deep Residual Learning[J]. Laser & Optoelectronics Progress, 2018, 55(3): 031004 Copy Citation Text show less
    Architecture of residual learning
    Fig. 1. Architecture of residual learning
    Architecture of the CNN
    Fig. 2. Architecture of the CNN
    Test images used in the experiment
    Fig. 3. Test images used in the experiment
    Results of different methods for test images under the same noise level. (a) Noise images; (b) Lee method; (c) Frost method; (d) BM3D method; (e) NL method; (f) CNN method
    Fig. 4. Results of different methods for test images under the same noise level. (a) Noise images; (b) Lee method; (c) Frost method; (d) BM3D method; (e) NL method; (f) CNN method
    Detail comparison of different denoising methods.(a) Lee method; (b) Frost method; (c) BM3D method; (d) NL method; (e) CNN method
    Fig. 5. Detail comparison of different denoising methods.(a) Lee method; (b) Frost method; (c) BM3D method; (d) NL method; (e) CNN method
    Denoising results of CNN method under different noise levels. (a) σ2=0.02; (b) σ2=0.04;(c) σ2=0.06; (d) σ2=0.08; (e) σ2=0.1
    Fig. 6. Denoising results of CNN method under different noise levels. (a) σ2=0.02; (b) σ2=0.04;(c) σ2=0.06; (d) σ2=0.08; (e) σ2=0.1
    (a) PSNR and (b) SSIM of different denoising methods versus noise level
    Fig. 7. (a) PSNR and (b) SSIM of different denoising methods versus noise level
    ImageLeeFrostBM3DNLCNN
    Lena26.4025.1420.8828.2230.59
    Baboon23.2122.3520.5422.3326.06
    Barbara23.5723.2021.6623.1628.45
    Boats25.4024.1420.3225.5028.65
    Peppers25.9724.8720.8728.4530.14
    Satellite25.5823.6220.0625.8328.38
    Average25.0223.8920.7225.5828.71
    Table 1. PSNR of different methods for test images under the same noise leveldB
    ImageLeeFrostBM3DNLCNN
    Lena0.6470.4960.4010.7330.846
    Baboon0.5640.4830.5170.4510.756
    Barbara0.6000.5100.5590.5870.855
    Boats0.6340.5100.4240.6470.784
    Peppers0.6420.4710.4160.7310.804
    Satellite0.5720.4250.3440.5720.719
    Average0.6100.4820.4440.6200.794
    Table 2. SSIM of different methods for test images under the same noise level
    NoiseimageLeeFrostBM3DNLCNN
    σ2=0.0227.7127.5825.0028.9532.99
    σ2=0.0426.3625.0820.8728.1930.66
    σ2=0.0625.3723.6018.8527.5329.25
    σ2=0.0824.5522.5217.5126.9728.19
    σ2=0.1024.0021.6216.4826.4227.18
    Average25.6024.0819.7427.6129.65
    Table 3. PSNR of different methods under different noise levelsdB
    NoiseimageLeeFrostBM3DNLCNN
    σ2=0.020.7360.6020.5580.7790.882
    σ2=0.040.6430.4950.4010.7300.843
    σ2=0.060.5830.4320.3320.6920.817
    σ2=0.080.5400.3910.2840.6620.795
    σ2=0.100.5010.3570.2450.6330.774
    Average0.6010.4550.3640.6990.822
    Table 4. SSIM of different methods under different noise levels