• Laser & Optoelectronics Progress
  • Vol. 58, Issue 20, 2015003 (2021)
Dai Chaodong1、2, Xu Guoliang2、*, Mao Jiao1、2, Gu Tong1、2, and Luo Jiangtao2
Author Affiliations
  • 1College of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China
  • 2Institute of Electronic Information and Network Engineering, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China
  • show less
    Cell phone screen texture background. (a) Defect-free image; (b) defect image
    Fig. 1. Cell phone screen texture background. (a) Defect-free image; (b) defect image
    Algorithm overall framework
    Fig. 2. Algorithm overall framework
    Denoising autoencoder structure
    Fig. 3. Denoising autoencoder structure
    Segmentation process
    Fig. 4. Segmentation process
    Residual image pixel distribution
    Fig. 5. Residual image pixel distribution
    Schematic of triangle method
    Fig. 6. Schematic of triangle method
    Pseudo-code graph of triangle threshold segmentation
    Fig. 7. Pseudo-code graph of triangle threshold segmentation
    Background reconstruction and defect segmentation results
    Fig. 8. Background reconstruction and defect segmentation results
    Comparison of defect segmentation results. (a) Defect images; (b) ground truth; (c) SVD; (d) U-Net; (e) MSCDAE; (f) proposed method
    Fig. 9. Comparison of defect segmentation results. (a) Defect images; (b) ground truth; (c) SVD; (d) U-Net; (e) MSCDAE; (f) proposed method
    CompositionNumber
    Point defect image831
    Line defect image757
    Block defect image968
    Defect-free image3000
    Table 1. Composition of the dataset
    Loss functionPSNR /dBSSIMLoss functionPSNR /dBSSIM
    L142.12590.9724SSIM+100L141.87970.9738
    L242.50810.9760SSIM+L242.28280.9795
    SSIM41.77160.978510SSIM+L240.10750.9785
    SSIM+L143.69480.9833100SSIM+L240.99730.9799
    10SSIM+L140.61200.9753SSIM+10L241.10220.9739
    100SSIM+L141.45170.9793SSIM+100L239.79730.9681
    SSIM+10L141.49880.9746
    Table 2. Reconstruction performance of different loss functions
    Number of layersPSNR /dBSSIM
    3-layer37.37690.9641
    4-layer41.94430.9754
    5-layer43.37540.9808
    6-layer41.92290.9742
    7-layer39.16760.9615
    Table 3. Reconstruction performance of different network layers
    AlgorithmPrecisonRecallF1-scoremIoU /%
    SVD[1]0.95070.86000.895984.15
    U-Net[8]0.95380.80170.846480.16
    MSCDAE[13]0.95170.92500.933389.10
    Proposed method0.97200.92470.944690.30
    Table 4. Comparison of experimental index
    Copy Citation Text
    Chaodong Dai, Guoliang Xu, Jiao Mao, Tong Gu, Jiangtao Luo. Cell Phone Screen Defect Segmentation Based on Unsupervised Network[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2015003
    Download Citation
    Category: Machine Vision
    Received: Nov. 25, 2020
    Accepted: Jan. 11, 2021
    Published Online: Oct. 14, 2021
    The Author Email: Xu Guoliang (xugl@cqupt.edu.cn)