Optics and Precision Engineering, Volume. 29, Issue 9, 2210(2021)

Dense irregular text detection based on multi-dimensional convolution fusion

Yue-bo MENG1...2, De-wang SHI1, Guang-hui LIU1,*, Sheng-jun XU1,2 and Dan JIN1 |Show fewer author(s)
Author Affiliations
  • 1College of Information and Control Engineering,Xi 'an University of Architecture and Technology, Xi'an70055, China
  • 2Guangzhou Artificial Intelligence and Digital Economy Laboratory, Guangzhou510000, China
  • show less
    Figures & Tables(15)
    Network basic structure
    TEM modular structure
    CFSi strategy structure
    Text candidate box generation process based on progressive expanded text kernel
    Accuracy and loss curves
    Experiential results of initial feature extraction and enhanced text feature extraction
    Experimental results of FPN text multidimensional feature extraction
    Experimental results of CFS module
    Text kernel S1, S2, S3 text instance segmentation and final detection results
    Detection result of DAST1500
    Detection result of ICDAR2015
    Detection result of CTW1500
    • Table 1. Experimental results of DAST1500

      View table
      View in Article

      Table 1. Experimental results of DAST1500

      MethedTEMCFSPrecisionRecallF-score
      EAST1369.255.861.8
      SegLink767.263.865.5
      TextSnake1473.672.172.8
      PSENet1578.876.277.4
      Our Method79.277.878.5
      Our Method79.378.178.7
      Our Method81.781.981.8
    • Table 2. Experimental results of ICDAR2015

      View table
      View in Article

      Table 2. Experimental results of ICDAR2015

      MethedTEMCFSPrecisionRecallF-score
      SegLink773.176.875.0
      EAST1383.673.578.2
      PSENet1581.579.780.6
      TextSnake1484.980.482.6
      Our Method81.780.180.9
      Our Method83.380.681.9
      Our Method84.480.783.0
    • Table 3. Experimental results of CTW1500

      View table
      View in Article

      Table 3. Experimental results of CTW1500

      MethedTEMCFSPrecisionRecallF-score
      SegLink742.340.040.8
      EAST1378.749.160.4
      TextSnake1467.985.375.6
      PSENet1580.675.678.0
      Our Method80.575.978.1
      Our Method81.776.078.7
      Our Method82.176.179.0
    Tools

    Get Citation

    Copy Citation Text

    Yue-bo MENG, De-wang SHI, Guang-hui LIU, Sheng-jun XU, Dan JIN. Dense irregular text detection based on multi-dimensional convolution fusion[J]. Optics and Precision Engineering, 2021, 29(9): 2210

    Download Citation

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

    Category: Information Sciences

    Received: Feb. 24, 2021

    Accepted: --

    Published Online: Nov. 22, 2021

    The Author Email: LIU Guang-hui (guanghui1@163.com)

    DOI:10.37188/OPE.20212909.2210

    Topics