Optical Instruments, Volume. 45, Issue 1, 8(2023)

Face image frontalization method for face expression analysis

Xuedian ZHANG, Zhongjun CHEN*, and Xiaofei QIN
Author Affiliations
  • School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • show less
    Figures & Tables(21)
    Flow chart of the method in this paper
    68-point landmark in face + another 3 points
    Landmark frontalization
    The change of average faces under different methods
    Face rough alignment
    Face deformation
    Face information processing
    Batch 1 experiments
    Batch 2 experiments
    Batch 3 experiments
    Batch 4 experiments
    Batch 5 experiments
    Ranking of the highest accuracy of various operations in a combined scenario
    • Table 1. Data set allocation

      View table
      View in Article

      Table 1. Data set allocation

      数据集训练集/张验证集/张总计/张
      RAF-DB7801338111182
      ExpW14195608620281
    • Table 2. Experimental design

      View table
      View in Article

      Table 2. Experimental design

      实验批次数据集网络模型信息比(a操作
      1RAFDBVGG16≤1简单裁剪、传统对齐、Hassner方法、变形、对称、融合
      2RAFDBVGG16<0.5传统对齐、变形、对称、融合
      3RAFDBVGG16≥0.5传统对齐、Hassner方法、 变形、对称、融合、变形增强、对称增强、融合增强
      4RAFDBResNet50≥0.5传统对齐、Hassner方法、变形、对称、融合
      5ExpWVGG16≥0.5传统对齐、Hassner方法、变形、对称、融合
    • Table 3. Batch 1 experiments: accuracy in combined scenarios

      View table
      View in Article

      Table 3. Batch 1 experiments: accuracy in combined scenarios

      操作Best_Acc/%Mean_Acc/%
      简单裁剪70.768.7
      传统对齐70.667.9
      Hassner方法67.765.8
      变形(本文方法)73.571.6
      对称(本文方法)74.272.5
      融合(本文方法)75.273.6
    • Table 4. Batch 2 experiments: accuracy in scenarios with large deflection angles

      View table
      View in Article

      Table 4. Batch 2 experiments: accuracy in scenarios with large deflection angles

      操作Best _Acc/%Mean_Acc/%
      传统对齐58.252.7
      变形(本文方法)70.066.0
      对称(本文方法)71.769.4
      融合(本文方法)70.467.9
    • Table 5. Batch 3 experiments: accuracy in scenarios with small deflection angles

      View table
      View in Article

      Table 5. Batch 3 experiments: accuracy in scenarios with small deflection angles

      操作Best _Acc/%Mean_Acc/%
      传统对齐69.066.1
      Hassner方法68.866.7
      变形(本文方法)74.973.6
      对称(本文方法)72.570.5
      融合(本文方法)74.372.7
      变形增强(本文方法)75.673.8
      对称增强(本文方法)75.673.4
      融合增强(本文方法)73.671.4
    • Table 6. Batch 4 experiments: accuracy in other model scenarios

      View table
      View in Article

      Table 6. Batch 4 experiments: accuracy in other model scenarios

      操作Best _Acc(%)Mean_Acc(%)
      传统对齐70.968.3
      Hassner方法69.465.4
      变形(本文方法)74.772.0
      对称(本文方法)71.869.9
      融合(本文方法)73.772.0
    • Table 7. Batch 5 experiments: accuracy in other dataset scenarios

      View table
      View in Article

      Table 7. Batch 5 experiments: accuracy in other dataset scenarios

      操作编号Best _Acc/%Mean_Acc/%
      传统对齐57.752.3
      Hassner方法58.953.5
      变形(本文方法)61.155.0
      对称(本文方法)60.055.2
      融合(本文方法)62.455.8
    • Table 8. The highest accuracy obtained by retraining the pre-trained model

      View table
      View in Article

      Table 8. The highest accuracy obtained by retraining the pre-trained model

      操作原图简单对齐Hassner方法粗对齐(本文方法)
      Best _Acc/%81.183.781.284.4
    Tools

    Get Citation

    Copy Citation Text

    Xuedian ZHANG, Zhongjun CHEN, Xiaofei QIN. Face image frontalization method for face expression analysis[J]. Optical Instruments, 2023, 45(1): 8

    Download Citation

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

    Category: APPLICATION TECHNOLOGY

    Received: Feb. 26, 2022

    Accepted: --

    Published Online: Mar. 20, 2023

    The Author Email: Zhongjun CHEN (203590797@st.usst.edu.cn)

    DOI:10.3969/j.issn.1005-5630.2023.001.002

    Topics