• Laser & Optoelectronics Progress
  • Vol. 57, Issue 22, 221019 (2020)
Ying Chen and Shuhui Gao*
Author Affiliations
  • School of Criminal Investigation and Forensic Science, People's Public Security University of China, Beijing 100038, China
  • show less
    DOI: 10.3788/LOP57.221019 Cite this Article Set citation alerts
    Ying Chen, Shuhui Gao. Forgery Numeral Handwriting Detection Based on Fire Module Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2020, 57(22): 221019 Copy Citation Text show less
    Fire Module structure
    Fig. 1. Fire Module structure
    Network structure. (a) FNNet structure; (b) AlexNet structure
    Fig. 2. Network structure. (a) FNNet structure; (b) AlexNet structure
    Part dataset examples. (a) Normal samples; (b) forgery samples
    Fig. 3. Part dataset examples. (a) Normal samples; (b) forgery samples
    ROC curves of six forged figures
    Fig. 4. ROC curves of six forged figures
    Comparison of test accuracy between FNNet and AlexNet
    Fig. 5. Comparison of test accuracy between FNNet and AlexNet
    Test accuracy of FNNet and other algorithms
    Fig. 6. Test accuracy of FNNet and other algorithms
    Layer nameOutput sizeFilter size/strides1×1e1×1e3×3Dimension
    Input image224×224×3
    Conv154×54×9611×11/434,944
    Maxpool127×27×963×3/2
    Fire227×27×2563212812844,320
    Fire327×27×38448192192104,880
    Fire427×27×38448192192111,024
    Fire527×27×51264256256188,992
    Maxpool513×13×5123×3/2
    Fire613×13×51264256256197,184
    Conv715×15×2561×1/1131,328
    Maxpool77×7×2563×3/2
    FC87×7×100012,545,000
    FC91×1×22002
    Table 1. Dimension parameters of FNNet
    Dataset346789
    Normal handwriting540593599599604597
    Forgery handwriting621612616614600614
    Table 2. Composition of dataset
    Parameter50010002000
    Test accuracy /%95.2896.3293.98
    Loss value0.140.110.25
    Train time /min131314
    Table 3. Classification and loss performances of number of neurons in fully connected layer
    ForgerynumeralTrainaccuracy /%LossvalueTestaccuracy /%
    399.600.0399.20
    496.400.1196.32
    698.400.0798.32
    798.400.0498.72
    899.200.0199.60
    997.800.0698.00
    Table 4. Results of training and testing of proposed network
    NetworkMean test accuracy /%Parameter quantity
    AlexNet95.3558,289,538
    FNNet98.3613,228,346
    Table 5. Parameter comparison of FNNet and AlexNet
    Ying Chen, Shuhui Gao. Forgery Numeral Handwriting Detection Based on Fire Module Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2020, 57(22): 221019
    Download Citation