OCR Util

Collection of utils to

• evaluation of OCR data for the masses
• generation of extended OCR-Evaluation Corpora
• generation of pair-wise Trainingdata for OCR-Backends

Requirements

• recent *nix-OS
• Python3.10+ Environment

Usage

Each section contains detailed usage help instructions:

# evaluation
ocr eval --help

# corpus management
ocr corpus --help

# slice image by image + input OCR
ocr slice --help

# render image + input OCR
ocr show --help

Data problems

Inconsistent OCR Groundtruth with empty texts (ALTO String elements missing CONTENT or PAGE without TextEquiv) or invalid geometrical coordinates (less than 3 points or even empty) will lead to evaluation errors if geometry must be respected.

Please note:
Invalid data files are tried(!) to be excluded from evaluation.

Evaluation Filter-Then-Aggregate

The evaluation CLI supports a single pre-aggregation filter using metadata extractors.

Example: keep only entries where MODS language is exactly German, then aggregate by publication century:

ocr eval <candidates> \
	--reference <groundtruth> \
	--mets-file <mets.xml> \
	--filter-by "mods:language=ger" \
	--aggregate-by "mods:dateIssued:century"

Multi-language filter values are interpreted as sets:

ocr eval <candidates> \
	--reference <groundtruth> \
	--mets-file <mets.xml> \
	--filter-by "mods:language=ger+eng" \
	--aggregate-by "mods:dateIssued:century"

Behavior:

• single filter value -> exact match (e.g. ger does not match ger+eng)
• multi-value filter -> all filter values must be present in any order
• entries missing the filter criterion are reported as WARNING and discarded

Development

Plattform: Intel(R) Core(TM) i5-6500 CPU@3.20GHz, 16GB RAM, Ubuntu 22.04 LTS, Python 3.10+

# clone local
git clone <repository-url> <local-dir>
cd <local-dir>

# enable virtual python 3 environment (linux)
# and update pip itself
python3.10 -m venv venv
. venv/bin/activate
python -m pip install -U pip

# install with dev dependencies
python -m pip install -e ".[dev,test]"

# run tests with coverage
python -m pytest --cov=src

# run tests faster (parallel, auto worker count)
python -m pytest -q -n auto

Contribution

Contributions, suggestions and proposals welcome!

License

Under terms of the MIT license.

NOTE: This software depends on packages that might be licensed under different terms.