/AFL

Implementation of "Robust offline handwritten character recognition through exploring writer-independent features under the guidance of printed data" (AFL)

Primary LanguagePython

Implementation of "Robust offline handwritten character recognition through exploring writer-independent features under the guidance of printed data" (AFL)

Paper

YapingZhang, Shan Liang, Shuai Nie, Wenju Liu, Shouye Peng, "Robust offline handwritten character recognition through exploring writer-independent features under the guidance of printed data", PR Letter 2018

Dependency

  • Please use python3, as we cannot guarantee its compatibility with python2.

  • Our code is based on Anaconda.

  • The version of Tensorflow we use is 1.10.1.

  • Other depencencies:

    pip install keras

Usage

  1. Clone the repo.

    git clone https://github.com/AprilYapingZhang/AFL.git
cd AFL

Using ready-made data

  1. Download the prepared data in hdf5 from Baidu Yun with passwd pf7f , to the repo root, and uncompress it.

    NOTE: For the raw CASIA-HWDB, built by the CASIA, are released for academic research free of cost under an agreement.

  2. Make sure the structure looks like the following:

    data/:
CASIA_HWDB_1.0_1.1_data
data/CASIA_HWDB_1.0_1.1_data:
norm_hand_pair_3755.hdf5  trn-HWDB1.0-1.1-3756-uint8.hdf5  tst-HWDB1.0-1.1-3756-uint8.hdf5

  3. Run model

    • Download the pretrain weights from Baidu Yun with passwd ie3j , to the repo root.

    • Run Baseline:

      python baseline.py --data_dir ./data

    • Run AFL model:

      python main.py --data_dir ./data --pretrain_weights ./pre_weights.hdf5

Bibtex

@article{zhang2018robust,
  title={Robust offline handwritten character recognition through exploring writer-independent features under the guidance of printed data},
  author={Zhang, Yaping and Liang, Shan and Nie, Shuai and Liu, Wenju and Peng, Shouye},
  journal={Pattern Recognition Letters},
  volume={106},
  pages={20--26},
  year={2018},
  publisher={Elsevier}
}

Acknowledgements

  • This code is built on keras.
  • The authors are grateful that Professor Cheng-lin Liu shared the CASIA-HWDB databases for our research.