2020-01-15 19:49:18 +01:00
|
|
|
# Binarization
|
2020-10-15 14:44:41 +02:00
|
|
|
|
2020-01-15 19:49:18 +01:00
|
|
|
> Binarization for document images
|
2019-12-10 13:00:05 +01:00
|
|
|
|
2023-04-13 10:50:06 +02:00
|
|
|
[](https://pypi.org/project/sbb-binarization/)
|
2023-04-13 22:33:56 +02:00
|
|
|
[](https://circleci.com/gh/qurator-spk/sbb_binarization)
|
|
|
|
|
[](https://github.com/qurator-spk/sbb_binarization/actions/workflows/test.yml)
|
2023-04-13 10:50:06 +02:00
|
|
|
|
2020-10-29 00:48:24 +01:00
|
|
|
## Examples
|
2020-10-29 00:45:55 +01:00
|
|
|
|
2020-10-29 00:48:03 +01:00
|
|
|
<img src="https://user-images.githubusercontent.com/952378/63592437-e433e400-c5b1-11e9-9c2d-889c6e93d748.jpg" width="180"><img src="https://user-images.githubusercontent.com/952378/63592435-e433e400-c5b1-11e9-88e4-3e441b61fa67.jpg" width="180"><img src="https://user-images.githubusercontent.com/952378/63592440-e4cc7a80-c5b1-11e9-8964-2cd1b22c87be.jpg" width="220"><img src="https://user-images.githubusercontent.com/952378/63592438-e4cc7a80-c5b1-11e9-86dc-a9e9f8555422.jpg" width="220">
|
2020-10-29 00:45:55 +01:00
|
|
|
|
2020-01-15 19:49:18 +01:00
|
|
|
## Introduction
|
2020-10-15 14:44:41 +02:00
|
|
|
|
2022-10-11 11:37:00 +02:00
|
|
|
This tool performs document image binarization using a trained ResNet50-UNet model.
|
2019-12-10 13:00:05 +01:00
|
|
|
|
2020-01-15 19:49:18 +01:00
|
|
|
## Installation
|
2020-10-15 14:44:41 +02:00
|
|
|
|
2020-10-23 19:26:51 +02:00
|
|
|
Clone the repository, enter it and run
|
2020-10-15 14:44:41 +02:00
|
|
|
|
|
|
|
|
`pip install .`
|
2020-01-15 19:49:18 +01:00
|
|
|
|
2020-01-16 15:56:32 +01:00
|
|
|
### Models
|
2020-10-15 14:44:41 +02:00
|
|
|
|
2023-04-12 22:55:12 +02:00
|
|
|
Pre-trained models in HDF5 format can be downloaded from here:
|
2020-10-15 14:44:41 +02:00
|
|
|
|
2020-01-16 15:47:52 +01:00
|
|
|
https://qurator-data.de/sbb_binarization/
|
2020-01-15 19:49:18 +01:00
|
|
|
|
2023-04-13 10:50:06 +02:00
|
|
|
We also provide models in Tensorflow SavedModel format via Huggingface and Github release assets:
|
2022-10-11 11:37:00 +02:00
|
|
|
|
|
|
|
|
https://huggingface.co/SBB/sbb_binarization
|
2023-04-13 10:50:06 +02:00
|
|
|
https://github.com/qurator-spk/sbb_binarization/releases
|
2022-10-11 11:37:00 +02:00
|
|
|
|
2023-04-12 22:55:12 +02:00
|
|
|
With [OCR-D](https://ocr-d.de/), you can use the [Resource Manager](Tensorflow SavedModel) to deploy models, e.g.
|
|
|
|
|
|
|
|
|
|
ocrd resmgr download ocrd-sbb-binarize "*"
|
|
|
|
|
|
|
|
|
|
|
2020-10-23 19:26:51 +02:00
|
|
|
## Usage
|
2020-10-15 14:44:41 +02:00
|
|
|
|
|
|
|
|
```sh
|
|
|
|
|
sbb_binarize \
|
2022-10-12 00:29:22 +02:00
|
|
|
-m <path to directory containing model files \
|
2020-10-23 19:26:51 +02:00
|
|
|
<input image> \
|
|
|
|
|
<output image>
|
2020-10-15 14:44:41 +02:00
|
|
|
```
|
2020-10-23 19:26:51 +02:00
|
|
|
|
2022-10-11 11:37:00 +02:00
|
|
|
Images containing a lot of border noise (black pixels) should be cropped beforehand to improve the quality of results.
|
2022-04-25 19:24:04 +02:00
|
|
|
|
2022-10-11 11:37:00 +02:00
|
|
|
### Example
|
2022-04-25 19:24:04 +02:00
|
|
|
|
2021-01-18 15:17:46 +01:00
|
|
|
|
2023-04-12 22:55:12 +02:00
|
|
|
sbb_binarize -m /path/to/model/ myimage.tif myimage-bin.tif
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
To use the [OCR-D](https://ocr-d.de/en/spec/cli) 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):
|
2023-04-13 10:50:06 +02:00
|
|
|
|
|
|
|
|
make model
|
|
|
|
|
make test
|
2023-04-13 22:33:56 +02:00
|
|
|
|
