OCR-D setup guide

OCR-D’s software is a modular collection of many projects (called modules) with many tools per module (called processors) that you can combine freely to achieve the workflow best suited for OCRing your content.

System requirements

Minimum system requirements

Installation

ocrd_all

ocrd_all is the main way to distribute and install the OCR-D software. If you want to produce OCR output from image data, this is what you need.

Installation: Docker or Native

There are two methods to install OCR-D:

1. Docker Installation of OCR-D using the prebuilt ocrd/all Docker images to install a module collection (recommended)
2. Native Installation of OCR-D using the ocrd_all git repository to install selected modules natively

We recommend using the prebuilt Docker images, since this does not require any changes to the host system besides installing Docker.

ocrd_all via Docker

Prerequisites

If you want to use the OCR-D-via-Docker solution, docker has to be installed.

After installing docker you may have to set up and start the docker daemon and add your user to the docker group:

# Start docker daemon at startup
sudo systemctl enable docker
# Add user to group 'docker'
sudo usermod -aG docker $USER

Please log out and log in again.

To test access to docker try the following command:

docker images

Now you should see an (empty) list of available images.

mini medi maxi

There are three versions of the ocrd/all Docker image: minimum, medium and maximum. They differ in which modules are included and hence the size of the image:

• minimum is comprised of the essential OCR-D components, with Tesseract and OCRopus as OCR engines.
• medium adds the Calamari OCR engine, as well as extra segmentation, pre- and postprocessing options.
• maximum includes all modules for best performance and full flexibility, but requires the most disk space.

We encourage the use of the relatively large but complete maximum image. The minimum or medium images should only be used when certain that none but the included OCR-D modules are needed.

Fetch Docker image

To fetch the maximum version of the ocrd/all Docker image:
(replace maximum accordingly if you want the minimum or medium version)

docker pull ocrd/all:maximum

Testing the Docker installation

To start, download and extract a document from the OCR-D GT Repo:

wget "https://ola-hd.ocr-d.de/api/export?id=21.11156/BFBAD520-65F4-430A-B4B2-C81A296C9E09&internalId=false" -O wundt_grundriss_1896.ocrd.zip
unzip wundt_grundriss_1896.ocrd.zip
cd data

Now, spin up the docker container:

docker run --workdir /data --volume $PWD:/data --rm -it ocrd/all bash

Your command line should start with something similar to:

root@c5d94e852de0:/data#

After spinning up the container, you can use the installation and call the processors the same way as in the native installation.

Alternatively, you can translate each command to a docker call.

Let’s segment the images in file group OCR-D-IMG from the zip file into regions, thereby creating a METS file group OCR-D-SEG-BLOCK-DOCKER):

ocrd-tesserocr-segment-region -I OCR-D-IMG -O OCR-D-SEG-BLOCK-DOCKER

When you are finished using OCR-D commands, use this command to stop using docker interactively:

exit

Updating Docker image

To update the Docker image to the latest version, just run the docker pull command:
(replace maximum accordingly if you use the minimum or medium version)

docker pull ocrd/all:maximum

Further reading

We recommend jumping to the section about installing models at the bottom of this page next. Alternatively, for instructions on how to proceed further with the processing of your data, please see the user guide. Make sure to also read the notes on translating native command line calls to docker calls.

ocrd_all natively

The ocrd_all project contains a sophisticated Makefile to install or compile prerequisites as necessary, set up a virtualenv, install the core software, install OCR-D modules and more. Detailed documentation can be found in its README.

Installation

There are some system requirements for ocrd_all.

You need to have make installed to make use of ocrd_all:

sudo apt install make

Clone the repository (still without submodules) and change into the ocrd_all directory:

git clone https://github.com/OCR-D/ocrd_all
cd ocrd_all

You should now be in a directory called ocrd_all.

It is easiest to install all the possible system requirements by calling make deps-ubuntu as root:

sudo make deps-ubuntu

This will install all system requirements.

Now you are ready for the final step which will actually install the OCR-D-Software.

You can either install

1. all the software at once with the all target (equivalent to the maximum Docker version),
2. modules individually by using an executable from that module as the target, or
3. a subset of modules by listing the project names in the OCRD_MODULES variable (equivalent to a custom selection of the medium Docker version):

make all                       # Installs all the software (recommended)

make ocrd-tesserocr-binarize   # Install ocrd_tesserocr which contains ocrd-tesserocr-binarize
make ocrd-cis-ocropy-binarize  # Install ocrd_cis  which contains ocrd-cis-ocropy-binarize

make all OCRD_MODULES="core ocrd_tesserocr ocrd_cis" # Will install only ocrd_tesserocr and ocrd_cis

(Custom choices for OCRD_MODULES and other control variables (cf. make help) can also be made permanent by writing them into local.mk.)

Note: Never run make all as root unless you know exactly what you are doing!

Installation is incremental, i.e. failed/interrupted attempts can be continued, and modules can be installed one at a time as needed.

Running make will also take care of cloning and updating all required submodules.

Especially running make all will take a while (between 30 and 60 minutes or more on slower machines). In the end, it should say that the last processor was installed successfully.

Having installed ocrd_all successfully, ocrd --version should give you the current version of OCR-D/core. Activate the virtual Python environment, which was created in the directory venv, before running any OCR-D command.

source venv/bin/activate
ocrd --version
ocrd, version 2.13.2 # your version should be 2.13.2 or later

Testing the native installation

For example, let’s fetch a document from the OCR-D GT Repo:

wget "https://ola-hd.ocr-d.de/api/export?id=21.11156/BFBAD520-65F4-430A-B4B2-C81A296C9E09&internalId=false" -O wundt_grundriss_1896.ocrd.zip
sudo unzip wundt_grundriss_1896.ocrd.zip
cd data

If you haven’t done it already, activate your venv:

# Activate the venv
source /path/to/ocrd_all/venv/bin/activate

Let’s segment the images in file group OCR-D-IMG from the zip file into regions (creating a first PAGE-XML file group OCR-D-SEG-BLOCK):

ocrd-tesserocr-segment-region -I OCR-D-IMG -O OCR-D-SEG-BLOCK

Updating the software

As ocrd_all is in active development, it is wise to regularly update the repository and its submodules:

git pull

This will refresh the local clone of ocrd_all with the changes in the official ocrd_all GitHub repository.

Now you can install the changes with

make all

This will run the installation process for all submodules which have been changed. In the end, it should say that the last processor was installed successfully. --version for the processors which have been changed should give you its current version.

Further reading

We recommend jumping to the section about installing models at the bottom of this page next. For instructions on how to process your own data, please see the user guide.

Individual installation (experts only)

For developing purposes it might be useful to install modules individually, either with Docker or natively. With all variants of individual module installation, it will be up to you to keep the repositories up-to-date and installed. We therefore discourage individual installation of modules and recommend using ocrd_all as outlined above..

All OCR-D modules follow the same interface and common design patterns. So once you understand how to install and use one project, you know how to install and use all of them.

Individual Docker container

This is the best option if you want full control over which modules you actually intend to use while still profiting from the simple installation of Docker containers.

You need to have Docker

Many OCR-D modules are also published as Docker containers to DockerHub. To find the Docker image for a module, replace the ocrd_ prefix with ocrd/:

docker pull ocrd/tesserocr  # Installs ocrd_tesserocr
docker pull ocrd/olena  # Installs ocrd_olena

Now you can test your installation.

Native installation

Installing each module into your system natively requires you to know and install all its dependencies first. That can be system packages (or even system package repositories) or Python packages.

To learn about system dependencies, consult the module’s README files. In contrast, Python dependencies should be resolved automatically by using the Python package manager pip.

NOTE

ocrd_tesserocr requires tesseract-ocr >= 4.1.0. But the Tesseract packages bundled with Ubuntu < 19.10 are too old. If you are on Ubuntu 18.04 LTS, please enable Alexander Pozdnyakov PPA, which has up-to-date builds of tesseract and its dependencies:

sudo add-apt-repository ppa:alex-p/tesseract-ocr
sudo apt-get update

Next subsections:

• For Python you also first need virtualenv. Then you have two options:
• installing via PyPI or
• installing via local git clone.

virtualenv

• Always install python modules into a virtualenv
• Never run pip/pip3 as root

First install Python 3 and venv:

sudo apt install python3 python3-venv

# If you haven't created the venv yet:
python3 -m venv ~/venv
# Activate the venv
source ~/venv/bin/activate

Once you have activated the virtualenv, you should see (venv) prepended to your shell prompt.

From PyPI

This is the best option if you want to use the stable, released version of individual modules.

However, many modules require a number of non-Python (system) packages. For the exact list of packages you need to look at the README of the module in question. (If you are not on Ubuntu >= 18.04, then your requirements may deviate from that.)

For example to install ocrd_tesserocr from PyPI:

sudo apt-get install git python3 python3-pip python3-venv libtesseract-dev libleptonica-dev tesseract-ocr-eng tesseract-ocr wget
pip3 install ocrd_tesserocr

Many ocrd modules conventionally contain a Makefile with a deps-ubuntu target that can handle calls to apt-get for you:

sudo make deps-ubuntu

Now you can test your installation.

From git

This is the best option if you want to change the source code or install the latest, unpublished changes.

git clone https://github.com/OCR-D/ocrd_tesserocr
cd ocrd_tesserocr
sudo make deps-ubuntu # or manually with apt-get
make deps             # or pip3 install -r requirements
make install          # or pip3 install .

If you intend to develop a module, it is best to install the module editable:

pip install -e .

This way, you won’t have to reinstall after making changes.

Now you can test your installation.

Installing models

Several processors in OCR-D need pretrained models you have to install beforehand.

Please consult our instruction on models to get more information on how to download and install them.