-
Notifications
You must be signed in to change notification settings - Fork 2
Home
Welcome to the PRL2018_WIN wiki!
This repository contains source code of the paper "Text-Independent Writer Identification using Convolutional Neural Network". Read paper here.
If you want to refer to our paper, please use the following citation:
Hung Tuan Nguyen, Cuong Tuan Nguyen, Takeya Ino, Bipin Indurkhya, Masaki Nakagawa, Text-independent writer identification using convolutional neural network, Pattern Recognition Letters, 2018, ISSN 0167-8655, https://doi.org/10.1016/j.patrec.2018.07.022.
For questions or more details, please contact us via email addresses: ntuanhung@gmail.com or nakagawa (at) cc.tuat.ac.jp
This project is supervised under Prof. Nakagawa and Prof. Bipin from 2016 to 2018.
The main contributors to this project are Takeya Ino (Master student, 2015-2017), Cuong Tuan Nguyen (Ph.D. student, 2014-2017) and Hung Tuan Nguyen (Ph.D. student, 2017-2020). All of us are from Nakagawa Laboratory, Tokyo University of Agriculture and Technology (TUAT).
Install the following packages via pip3: numpy, tqdm, Pillow, pickle, gzip, scipy, tensorflow-gpu
We developed and evaluated our source code on Tensorflow 1.7.0 with Python 3.6.5.
Inside images folder, all image file should be named by <writer_id>_<image_name> such as 0_1234. All images should be resized to squared and unique size, for example, 64-by-64.
In configs folder, there are different configuration files which are used to train/evaluate model. Each configuration has three text files for train/valid/test sets.
train-files_<number_of_writers>users_<number_of_patterns_per_writer>patPerUser_SAME_<database_name>.txt
valid-files_<number_of_writers>users_<number_of_patterns_per_writer>patPerUser_SAME_<database_name>.txt
test-files_<number_of_writers>users_<number_of_patterns_per_writer>patPerUser_SAME_<database_name>.txt
Each configuration file has <number_of_writers> lines. For every line, the first value is <writer_id> which is followed by <number_of_patterns_per_writer> values represented for <image_name>.
Read more in this page