Forward Implementation of Fast and Compact CNN for Offline HCCR

This project is a forward model of a fast and compact convolutional neural network for offline handwritten Chinese chracter recognition(HCCR). For more information, please see the paper: "Xuefeng Xiao, Lianwen Jin∗, Yafeng Yang, Weixin Yang, Jun Sun, Tianhai Chang, Building Fast and Compact Convolutional Neural Networks for Offline Handwritten Chinese Character Recognition, Pattern Recognition, vol.72, no. 72-81, 2017.

Project components

Image -- contains 1000 gray-valued character samples for test, which have the size of 96*96.

ImageName_Label.txt -- store the names of the test samples and their corresponding labels.

src -- store the code and the parameters of our model

Makefile -- link the MKL and OpenCV library for code compilation.

run. sh -- the shell script for running the project; note that the OMP_NUM_THREADS should be set to 1 to acquired accurate run-time.

Usage

Prior to compilation, you need to install MKL and OpenCV, and modify Makefile if needed. After that, execute "run. sh" file and perform test.

Experiment result

Our forward model obtains an accuracy of 97.09% on the ICDAR 2013 offline HCCR competition datast and consumes a average run-time of 9.72ms for every sample. The experiment is carried out on a single desktop PC, equipped with an Intel® Core™ i7-6700 CPU @ 3.40GHz × 8, 16GB RAM, ubuntu 14.04 LTS operating system. All experiment are executed in the single-thread mode, without GPU acceleration. The size of our model parameters is only 2.34MB. Owing to the memory limitation, we just offer 1000 images from the competition dataset in our provided project.

Citation

Please cite our paper if it helps your research:

@article{xiao2017building,
  author = {Xuefeng Xiao, Lianwen Jin, Yafeng Yang, Weixin Yang, Jun Sun, Tianhai Chang},
  title = {Building Fast and Compact Convolutional Neural Networks for Offline Handwritten Chinese Character Recognition},
  journal={Pattern Recognition},
  volume={72},
  pages={72--81},
  year = {2017},
}