sbb_binarization

Document Image Binarization

Installation

Python 3.8-3.11 with Tensorflow <2.13 are currently supported. While newer versions might also work, we currently don't test this.

You can either install from PyPI via

pip install sbb-binarization

or clone the repository, enter it and install (editable) with

git clone git@github.com:qurator-spk/sbb_binarization.git
cd sbb_binarization; pip install -e .

Alternatively, download the prebuilt image from Dockerhub:

docker pull ocrd/sbb_binarization

Models

Pre-trained models can be downloaded from the locations below. We also provide models and model cards on 🤗

Version	Format	Download
2021-03-09	SavedModel	https://github.com/qurator-spk/sbb_binarization/releases/download/v0.0.11/saved_model_2021_03_09.zip
2021-03-09	HDF5	https://qurator-data.de/sbb_binarization/2021-03-09/models.tar.gz
2020-01-16	SavedModel	https://github.com/qurator-spk/sbb_binarization/releases/download/v0.0.11/saved_model_2020_01_16.zip
2020-01-16	HDF5	https://qurator-data.de/sbb_binarization/2020-01-16/models.tar.gz

With OCR-D, you can also use the Resource Manager, e.g.

ocrd resmgr download ocrd-sbb-binarize "*"

Usage

sbb_binarize \
  -m <path to directory containing model files> \
  <input image> \
  <output image>

Note: the output image MUST use either .tif or .png as file extension to produce a binary image. Input images can also be JPEG.

Images containing a lot of border noise (black pixels) should be cropped beforehand to improve the quality of results.

Example

sbb_binarize -m /path/to/model/ myimage.tif myimage-bin.tif

To use the OCR-D interface:

ocrd-sbb-binarize -I INPUT_FILE_GRP -O OCR-D-IMG-BIN -P model default

Testing

For simple smoke tests, the following will

• download models

• download test data

• run the OCR-D wrapper (on page and region level):

    make models
    make test
  

How to cite

If you find this tool useful in your work, please consider citing our paper:

@inproceedings{hip23rezanezhad2,
author    = {Vahid Rezanezhad and Konstantin Baierer and Clemens Neudecker},
editor    = {Apostolos Antonacopoulos and Christian Clausner and Maud Ehrmann and Kai Labusch and Clemens Neudecker},
title     = {A hybrid CNN-Transformer Model for Historical Document Image Binarization},
booktitle = {Proceedings of the 7th International Workshop on Historical Document Imaging and Processing {HIP} 2023, 
             San José, CA, USA, August 26, 2023},
year      = {2023},
url       = {https://doi.org/10.1145/3604951.3605508}
}