README.md

🐬DOLPHIN

Online Writer Retrieval with Chinese Handwritten Phrases: A Synergistic Temporal-Frequency Representation Learning Approach

IEEE Transactions of Information Forensics and Security (TIFS), 2024

⭐Official code of the DOLPHIN model and the release of the OLIWER dataset.

🌊Introduction

DOLPHIN is an online writer retrieval model, designed to retrieve all online handwriting samples of a specific writer. It synergizes temporal and frequency learning to extract discriminative feature representations for online handwriting.

The model architecture of DOLPIHN

🌏Environment

git clone https://github.com/SCUT-DLVCLab/DOLPHIN.git
conda create -n dolphin python=3.8.16
conda activate dolphin
pip install -r requirements.txt

⚒️Data Preparation

Download the three subsets: CASIA-OLHWDB2, DCOH-E, and SCUT-COUCH2009 using the following links:

• Baidu Cloud
• Google Drive

Unzip the .zip archives using the following commands:

unzip OLHWDB2.zip -d .
unzip DCOH-E.zip -d .
unzip COUCH09.zip -d .

The directory should look like this:

data-raw
├── COUCH09
│   ├── 001
│   └── ...
├── DCOH-E
│   ├── dcoh-e313
│   └── ...
└── OLHWDB2
    ├── 001
    └── ...

Then run preprocess.py for data preprocessing:

python preprocess.py --dataset olhwdb2
python preprocess.py --dataset dcohe
python preprocess.py --dataset couch

The preprocessed data will be saved at the data folder.

Then run the divide.py to merge the three subsets into the OLIWER dataset and divide the data into training and testing parts.

python divide.py --divide
python divide.py --extract

Now the data should be all preprocessed. The final data directory should look like:

data
├── COUCH09
│   └── COUCH09.pkl
├── DCOH-E
│   └── DCOH-E.pkl
├── OLHWDB2
│   └── OLHWDB2.pkl
└── OLIWER
    ├── split.json
    ├── test.pkl
    ├── test-tf.pkl
    ├── train.pkl
    └── train-tf.pkl

🚀Test

python test.py --weights weights/model.pth

📑Citation

@ARTICLE{10746457,
  author={Zhang, Peirong and Jin, Lianwen},
  journal={IEEE Transactions on Information Forensics and Security}, 
  title={{Online Writer Retrieval With Chinese Handwritten Phrases: A Synergistic Temporal-Frequency Representation Learning Approach}}, 
  year={2024},
  volume={19},
  number={},
  pages={10387-10399},

☎️Cotact

Peirong Zhang: eeprzhang@mail.scut.edu.cn

🌴Copyright

Copyright 2024, Deep Learning and Vision Computing (DLVC) Lab, South China China University of Technology. http://www.dlvc-lab.net.