Improving Document Binarization via Adversarial Noise-Texture Augmentation (ICIP 2019)
This repository contains the full source codes and instructions to use the codes on the datasets described in the paper. The paper re-visits the binarization problem by introducingan adversarial learning approach.
In this paper, we propose a two-stage network that first learns to augment the document images by using neural style transfer technique. For this purpose, we construct a Texture Augmentation Network that transfers the texture element of a degraded reference document image to a clean binary image.
In this way, the network creates multiple versions of the same textual content with various noisy textures, enlarging the available document binarization datasets. At last, the newly generated images are passed through a Binarization network to get back the clean version.
If you find this code useful in your research, please consider citing:
@article{bhunia2018improving,
title={Improving Document Binarization via Adversarial Noise-Texture Augmentation},
author={Bhunia, Ankan Kumar and Bhunia, Ayan Kumar and Sain, Aneeshan and Roy, Partha Pratim},
journal={arXiv preprint arXiv:1810.11120},
year={2018}
}