• Acta Optica Sinica
  • Vol. 38, Issue 8, 0815024 (2018)
Sen Wang*, Xing Wu*, Yinhui Zhang, and Qing Chen
Author Affiliations
  • Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming, Yunnan 650500, China
  • show less
    DOI: 10.3788/AOS201838.0815024 Cite this Article Set citation alerts
    Sen Wang, Xing Wu, Yinhui Zhang, Qing Chen. Surface Crack Segmentation Based on Multi-Scale Wavelet Transform and Structured Forest[J]. Acta Optica Sinica, 2018, 38(8): 0815024 Copy Citation Text show less
    Flowchart of SFW method
    Fig. 1. Flowchart of SFW method
    Flowchart of modulus maxima edge detection method with wavelet semi-reconfiguration
    Fig. 2. Flowchart of modulus maxima edge detection method with wavelet semi-reconfiguration
    13 feature channels of wall crack
    Fig. 3. 13 feature channels of wall crack
    Feature vectors and structured labels
    Fig. 4. Feature vectors and structured labels
    Qualitative comparisons of wavelets with traditional detection methods. (a) Original image; (b) GT; (c) hrbio1.1-1; (d) rbior1.1-2; (e) sym2-1; (f) coif1-1; (g) dyadic-2; (h) dmey-1; (i) Prewitt; (j) Sobel; (k) Robert; (l) Canny; (m) Log
    Fig. 5. Qualitative comparisons of wavelets with traditional detection methods. (a) Original image; (b) GT; (c) hrbio1.1-1; (d) rbior1.1-2; (e) sym2-1; (f) coif1-1; (g) dyadic-2; (h) dmey-1; (i) Prewitt; (j) Sobel; (k) Robert; (l) Canny; (m) Log
    ROC curves and RPFM bars of six wavelets and six other methods. (a) ROC curves; (b) RPFM bars
    Fig. 6. ROC curves and RPFM bars of six wavelets and six other methods. (a) ROC curves; (b) RPFM bars
    Quantitative comparisons of SFW classifier in train and validation. (a) nSample; (b) nCell; (c) normRad; (d) chSmooth; (e) simSmooth; (f) imWidth; (g) gtWidth; (h) fracFtrs; (i) maxDepth; (j) minChild; (k) sharpen; (l) nTree
    Fig. 7. Quantitative comparisons of SFW classifier in train and validation. (a) nSample; (b) nCell; (c) normRad; (d) chSmooth; (e) simSmooth; (f) imWidth; (g) gtWidth; (h) fracFtrs; (i) maxDepth; (j) minChild; (k) sharpen; (l) nTree
    Qualitative comparisons of different methods. (a) Original image 1; (b) GT1; (c) hrbio1.1-1; (d) rbior1.1-2; (e) dyadic-2; (f) sym2-1; (g) coif1-1; (h) dmey-1; (i) original image 2; (j) GT2; (k) SFW-M (hrbio1.1-1); (l) SFW-1; (m) SFD-M; (n) SFD-1; (o) FCN-8s; (p) MDW Ncut
    Fig. 8. Qualitative comparisons of different methods. (a) Original image 1; (b) GT1; (c) hrbio1.1-1; (d) rbior1.1-2; (e) dyadic-2; (f) sym2-1; (g) coif1-1; (h) dmey-1; (i) original image 2; (j) GT2; (k) SFW-M (hrbio1.1-1); (l) SFW-1; (m) SFD-M; (n) SFD-1; (o) FCN-8s; (p) MDW Ncut
    ROC curve and RPFM bars of 11 methods. (a) ROC curves; (b) RPFM bars
    Fig. 9. ROC curve and RPFM bars of 11 methods. (a) ROC curves; (b) RPFM bars
    Quantitative comparisons of 5 methods. (a) Original image; (b) GT; (c) SFW; (d) FCN-8s; (e) SFD; (f) Canny; (g) original image; (h) GT; (i) SFW; (j) FCN-8s; (k) SFD; (l) Canny
    Fig. 10. Quantitative comparisons of 5 methods. (a) Original image; (b) GT; (c) SFW; (d) FCN-8s; (e) SFD; (f) Canny; (g) original image; (h) GT; (i) SFW; (j) FCN-8s; (k) SFD; (l) Canny
    ROC curves of two types of images with five methods. (a) First type; (b) second type
    Fig. 11. ROC curves of two types of images with five methods. (a) First type; (b) second type
    MethodRPFMAETime /s
    hrbio1.10.78620.61520.64770.10040.5107
    rbio1.10.78090.61370.64560.11740.5547
    dyadic0.76340.60800.63800.10901.2804
    sym20.77950.61330.64500.12000.5231
    coif10.77410.61150.64270.10240.6539
    dmey-10.75430.60510.63410.15130.8794
    SFW-10.76190.60470.63720.24830.1220
    SFD- M0.77520.61160.64290.04870.3636
    SFD-10.76050.60670.63640.04970.0841
    FCN-8s0.77070.61080.64150.02141.3646
    M Ncut0.72280.59400.61950.10342.0019
    Table 1. Average comparisons of 11 methods with 5 quantitative methods
    Image typeParameterSFWFCN-8sSFDCanny
    Crack images ofthe unevenillumination surfaceR0.85050.80090.79610.7813
    P0.63060.61630.61490.6104
    F0.67070.65090.64900.6429
    MAE0.01350.00440.01630.0093
    Time /s0.75310.27370.28140.1165
    Crack imagesof the contaminatedsurfaceR0.83170.66100.81800.7347
    P0.62550.56980.62150.5957
    F0.66340.58850.65800.6229
    MAE0.00930.00560.00890.0094
    Time / s0.75310.27370.28140.1165
    Table 2. Quantitative comparisons of five methods
    Sen Wang, Xing Wu, Yinhui Zhang, Qing Chen. Surface Crack Segmentation Based on Multi-Scale Wavelet Transform and Structured Forest[J]. Acta Optica Sinica, 2018, 38(8): 0815024
    Download Citation