Tools to preprocess and segment scanned images for OCR-D
Requires Python >= 3.6.
-
Create a new
venvunless you already have onepython3 -m venv venv -
Activate the
venvsource venv/bin/activate -
To install from source, get GNU make and do:
make installThere are also prebuilds available on PyPI:
pip install ocrd_anybaseocr
(This will install both PyTorch and TensorFlow, along with their dependents.)
All tools, also called processors, abide by the CLI specifications for OCR-D, which roughly looks like:
ocrd-<processor-name> [-m <path to METs input file>] -I <input group> -O <output group> [-p <path to parameter file>]* [-P <param name> <param value>]*
For each page (or sub-segment), this processor takes a scanned colored / gray scale document image as input and computes a binarized (black and white) image.
Implemented via rule-based methods (percentile based adaptive background estimation in Ocrolib).
ocrd-anybaseocr-binarize -I OCR-D-IMG -O OCR-D-BIN -P operation_level line -P threshold 0.3
For each page (or sub-segment), this processor takes a document image as input and computes the skew angle of that. It also annotates a deskewed image.
The input images have to be binarized for this module to work.
Implemented via rule-based methods (binary projection profile entropy maximization in Ocrolib).
ocrd-anybaseocr-deskew -I OCR-D-BIN -O OCR-D-DESKEW -P maxskew 5.0 -P skewsteps 20 -P operation_level page
For each page, this processor takes a document image as input and computes the border around the page content area (i.e. removes textual noise as well as any other noise around the page frame). It also annotates a cropped image.
The input image does not need to be binarized, but should be deskewed for the module to work optimally.
Implemented via rule-based methods (gradient-based line segment detection and morphology based textline detection).
ocrd-anybaseocr-crop -I OCR-D-DESKEW -O OCR-D-CROP -P rulerAreaMax 0 -P marginLeft 0.1
For each page, this processor takes a document image as input and computes a morphed image which will make the text lines straight if they are curved.
The input image has to be binarized for the module to work, and should be cropped and deskewed for optimal quality.
Implemented via data-driven methods (neural GAN conditional image model trained with pix2pixHD/Pytorch).
ocrd resmgr download ocrd-anybaseocr-dewarp '*'
ocrd-anybaseocr-dewarp -I OCR-D-CROP -O OCR-D-DEWARP -P resize_mode none -P gpu_id -1
For each page, this processor takes a document image as an input and computes two images, separating the text and non-text parts.
The input image has to be binarized for the module to work, and should be cropped and deskewed for optimal quality.
Implemented via data-driven methods (neural pixel classifier model trained with Tensorflow/Keras).
ocrd resmgr download ocrd-anybaseocr-tiseg '*'
ocrd-anybaseocr-tiseg -I OCR-D-DEWARP -O OCR-D-TISEG -P use_deeplr true
For each page, this processor takes the raw document image as an input and computes a text region segmentation for it (distinguishing various types of text blocks).
The input image need not be binarized, but should be deskewed for the module to work optimally.
Implemented via data-driven methods (neural Mask-RCNN instance segmentation model trained with Tensorflow/Keras).
ocrd resmgr download ocrd-anybaseocr-block-segmentation '*'
ocrd-anybaseocr-block-segmentation -I OCR-D-TISEG -O OCR-D-BLOCK -P active_classes '["page-number", "paragraph", "heading", "drop-capital", "marginalia", "caption"]' -P min_confidence 0.8 -P post_process true
For each page (or region), this processor takes a cropped document image as an input and computes a textline segmentation for it.
The input image should be binarized and deskewed for the module to work.
Implemented via rule-based methods (gradient and morphology based line estimation in Ocrolib).
ocrd-anybaseocr-textline -I OCR-D-BLOCK -O OCR-D-LINE -P operation_level region
For the whole document, this processor takes all the cropped page images and their corresponding text regions as input and computes the logical structure (page types and sections).
The input image should be binarized and segmented for this module to work.
Implemented via data-driven methods (neural Inception-V3 image classification model trained with Tensorflow/Keras).
ocrd resmgr download ocrd-anybaseocr-layout-analysis '*'
ocrd-anybaseocr-layout-analysis -I OCR-D-LINE -O OCR-D-STRUCT
To test the tools under realistic conditions (on OCR-D workspaces), download OCR-D/assets. In particular, the code is tested with the dfki-testdata dataset.
To download the data:
make assets
To run module tests:
make test
To run processor/workflow tests:
make cli-test
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.