• Laser & Optoelectronics Progress
  • Vol. 56, Issue 24, 241501 (2019)
Hengjie Yang, Zheng Yan, Zongling Wu, Dingbang Fang, and Fang Duan*
Author Affiliations
  • College of Information Science and Engineering, Huaqiao University, Xiamen, Fujian 361021, China
  • show less
    DOI: 10.3788/LOP56.241501 Cite this Article Set citation alerts
    Hengjie Yang, Zheng Yan, Zongling Wu, Dingbang Fang, Fang Duan. Extraction Method of Interest Text in Image Based on Recurrent Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(24): 241501 Copy Citation Text show less
    Example of name entity recognition
    Fig. 1. Example of name entity recognition
    LSTM network unit
    Fig. 2. LSTM network unit
    Structure of forward long short time memory network
    Fig. 3. Structure of forward long short time memory network
    Structure of BLSTM network
    Fig. 4. Structure of BLSTM network
    Structure of CRF network
    Fig. 5. Structure of CRF network
    Structure of BLSTM-CRFs model
    Fig. 6. Structure of BLSTM-CRFs model
    Samples of text data and label generated in IDTRAIN and IDVAL. (a) Sample a; (b) sample b
    Fig. 7. Samples of text data and label generated in IDTRAIN and IDVAL. (a) Sample a; (b) sample b
    Samples of images in YNIDREAL
    Fig. 8. Samples of images in YNIDREAL
    Accuracy of six entities on IDVAL. (a) F1-score; (b) P value; (c) R value
    Fig. 9. Accuracy of six entities on IDVAL. (a) F1-score; (b) P value; (c) R value
    Test results on YNIDREAL dataset. (a) Text detection results; (b) text recognition results; (c) result of interest text extraction using BLSTM-CRF model; (d) result of interest text extraction using CRF model
    Fig. 10. Test results on YNIDREAL dataset. (a) Text detection results; (b) text recognition results; (c) result of interest text extraction using BLSTM-CRF model; (d) result of interest text extraction using CRF model
    ItemDataset categoryDataset typeDataset size
    TrainIDTRAINText500
    ValidationIDVALText100
    TestYNIDREALImage61
    Table 1. Distribution of experimental data set
    EntityCRFBLSTM-CRF
    P /%R /%F1 /%P /%R /%F1 /%
    Name75.0068.8571.7986.8986.8986.89
    Gender96.6795.0895.8796.7296.7296.72
    Nation95.0093.4494.2193.4493.4493.44
    Birth90.1690.1690.1691.8091.8091.80
    Address90.4893.4491.9493.6596.7295.16
    Idnum92.0695.0893.5590.4893.4491.94
    Average89.9089.3489.5992.1693.1792.66
    Table 2. System performances of CRF and BLSTM-CRF models
    ModelSucceednumberFailnumberSpeed /(image·s-1)Testaccuracy /%
    OCRExtraction
    CRF44170.179772.13
    BLSTM-CRF5470.178288.52
    Table 3. Test results of integrity of interest text extraction
    Hengjie Yang, Zheng Yan, Zongling Wu, Dingbang Fang, Fang Duan. Extraction Method of Interest Text in Image Based on Recurrent Neural Network[J]. Laser & Optoelectronics Progress, 2019, 56(24): 241501
    Download Citation