Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 6, 905(2025)

Arbitrary shape text detection method based on Fourier centerline prediction

Kun BAI1, Zhe WANG2, Long MA1, Yao XUE2, Guodong LI3, Tian YAN3, and Xiaotian WANG3、*
Author Affiliations
  • 1Xi'an Modern Control Technology Research Institute, Xi'an 710065, China
  • 2School of Information and Communications Engineering, Xi'an Jiaotong University, Xi'an 710049, China
  • 3Unmanned System Research Institute, Northwestern Polytechnical University, Xi'an 710072, China
  • show less
    Figures & Tables(9)
    Example diagram of centerline modeling
    Process of using Fourier circles to draw text centerlines
    Process of the proposed FCPNet
    Comparison of the detection quality of the method in this paper with TextRay[11], ABCNet[12], TESTR[6] and DPText[7] on selected challenging images from the CTW1500 dataset. Among them, TESTR and DPText are the latest Transformer-based methods.
    • Table 1. Ablation experiments

      View table
      View in Article

      Table 1. Ablation experiments

      TCRASFCTW1500
      P/%R/%F/%
      --75.269.672.3
      -77.475.976.6
      -83.070.976.5
      87.380.183.6
    • Table 2. Verification of the generalization ability using different amounts of training data on the CTW1500 dataset

      View table
      View in Article

      Table 2. Verification of the generalization ability using different amounts of training data on the CTW1500 dataset

      数据算法P/%R/%F/%
      50%TextRay1175.577.876.6
      ABCNet1271.180.675.6
      FCPNet81.875.478.5
      25%TextRay1167.974.971.2
      ABCNet1270.075.572.7
      FCPNet78.570.874.5
    • Table 3. Validation of generalization ability using different datasets

      View table
      View in Article

      Table 3. Validation of generalization ability using different datasets

      方法CTW1500→TotalTextTotalText→CTW1500
      P/%R/%F/%P/%R/%F/%
      CTD-TLOC1342.525.932.237.450.543.0
      TFE-PRPA-BCTS1455.531.039.835.157.843.6
      SADA(Res50-FPN)857.132.641.536.756.244.4
      FCPNet55.535.943.646.558.951.9
    • Table 4. Comparison of related methods on CTW1500 and TotalText dataset

      View table
      View in Article

      Table 4. Comparison of related methods on CTW1500 and TotalText dataset

      方法Ext.CTW1500TotalText
      P/%R/%F/%FPSP/%R/%F/%FPS
      FCPNet87.380.183.615.489.075.081.412.6
      TextSnake1585.367.975.61.182.774.578.41.1
      TextField379.883.081.4681.279.980.66
      SAST1681.281.781.527.685.675.580.2-
      PSENet1779.784.882.23.978.084.080.93.9
      TextRay1180.482.881.6-83.577.980.6-
      ABCNet1278.584.481.417.9--78.417.9
      PSENet1775.680.678.03.981.875.178.33.9
      ATRR1880.280.280.11080.976.278.510
      TFE-PRPA-BCTS1482.581.882.2-83.574.778.9-
    • Table 5. Performance comparison on the subset of highly curved text instances

      View table
      View in Article

      Table 5. Performance comparison on the subset of highly curved text instances

      方法P/%R/%F/%
      TextRay1185.167.275.1
      ABCNet1288.362.673.2
      TESTR686.571.778.4
      DPText792.469.379.2
      FCPNet83.676.379.8
    Tools

    Get Citation

    Copy Citation Text

    Kun BAI, Zhe WANG, Long MA, Yao XUE, Guodong LI, Tian YAN, Xiaotian WANG. Arbitrary shape text detection method based on Fourier centerline prediction[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(6): 905

    Download Citation

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

    Category:

    Received: Feb. 14, 2025

    Accepted: --

    Published Online: Jul. 14, 2025

    The Author Email: Xiaotian WANG (18710993786@163.com)

    DOI:10.37188/CJLCD.2025-0033

    Topics