Acta Optica Sinica, Volume. 41, Issue 4, 0410001(2021)

Retinal Blood Vessel Segmentation Based on Multi-Scale Wavelet Transform Fusion

Feng Tian*, Ying Li, and Jing Wang
Author Affiliations
  • College of Communication and Information Engineering, Xi'an University of Science and Technology, Xi'an, Shaanxi 710054, China
  • show less
    Figures & Tables(16)
    Flow chart of multi-scale blood vessel segmentation
    Preprocessed images. (a) Original color image; (b) green channel image; (c) contrast enhanced image by CLAHE
    Gradient amplitudes when σ=1,2,4 pixel. (a) σ=1 pixel; (b) σ=2 pixel; (c) σ=4 pixel
    Maximum principal curvature images when σ=1,2,4 pixel. (a) σ=1 pixel; (b) σ=2 pixel; (c) σ=4 pixel
    Results of post-processing for contour feature image. (a) Laplacian filtered image; (b) energy significant mapping image; (c) image obtained by mathematical morphology processing
    Results of post-processing for detail feature image. (a) Energy significant mapping image; (b) image obtained by mathematical morphology processing
    Fusion process of wavelet transform
    Fusion image with wavelet transform when σ=1 pixel
    Image obtained by multi-scale vascular detection
    Comparison of detail images obtained by multi-scale blood vessel detection. (a) Original color image; (b) contour feature, (c) detail feature, and (d) detail image of wavelet transform fusion at σ=1 pixel; detail images obtained by blood vessel detection when (e) σ=2 pixel and (f) σ=4 pixel; (g) detail image obtained by multi-scale blood vessel detection
    Comparison of blood vessel segmentation effects of retinal images. (a) Original color fundus retinal images; (b) gold standard images; (c) results of single scale vascular segmentation; (d) results of multi-scale vascular segmentation
    1D cross-section of middle row of marked subarea in the third image in Figs. 11 (c) and (d)
    Blood vessel segmentation results by fusing contour information with detail information at multi-scale or single-scale frames. (a) Results of fusion at multi-scale frame; (b) details of gold standard image; (c) results of fusion at single-scale frame; (d) results of fusion at multi-scale frame
    Results of blood vessel segmentation with and without fusion of contour information and detail information at multiple scales. (a) Without fusion of contour information and detail information; (b) with fusion of contour information and detail information; (c) detailed drawing marked in Fig. 14(a); (d) detailed drawing marked in Fig. 14(b)
    • Table 1. Calculation method for evaluation indexes of vascular segmentation of fundus retinal image

      View table

      Table 1. Calculation method for evaluation indexes of vascular segmentation of fundus retinal image

      Evaluation indexDescription
      AccAcc=(TP+TN)/(TP+FP+TN+FN)
      SeSe=TP/(TP+FN)
      SpSp=TN/(TN+FP)
    • Table 2. Performance comparison of segmentation methods for fundus retinal blood vessels

      View table

      Table 2. Performance comparison of segmentation methods for fundus retinal blood vessels

      MethodAccSeSp
      Method in Ref. [7]0.93400.70600.9693
      Method in Ref. [26]0.94200.67700.9810
      Method in Ref. [27]0.94630.71890.9793
      Method in Ref. [9]0.94650.71650.9801
      Proposed method0.95820.70860.9806
    Tools

    Get Citation

    Copy Citation Text

    Feng Tian, Ying Li, Jing Wang. Retinal Blood Vessel Segmentation Based on Multi-Scale Wavelet Transform Fusion[J]. Acta Optica Sinica, 2021, 41(4): 0410001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Aug. 24, 2020

    Accepted: Sep. 30, 2020

    Published Online: Feb. 25, 2021

    The Author Email: Feng Tian (tianfeng@xust.edu.cn)

    DOI:10.3788/AOS202141.0410001

    Topics