ocrd_segment
This repository aims to provide a number of OCR-D compliant processors for layout analysis and evaluation.
Installation
In your Python virtual environment, run:
pip install ocrd_segment
Usage
Contains processors for various tasks:
- exporting segment images (including results from preprocessing like cropping/masking, deskewing, dewarping or binarization) along with polygon coordinates and metadata:
- importing layout segmentations from other formats:
- post-processing or repairing layout segmentations:
- ocrd-segment-repair (validity and consistency of all coordinates; also, for regions, reduce overlaps/redundancy between neighbours, and/or simplify polygons, and/or shrink to the alpha shape of foreground contours)
- ocrd-segment-project (remake segment coordinates into the concave hull / alpha shape of their constituents)
- ocrd-segment-replace-original (rebase all segments on cropped+deskewed border frame as new full page)
- ocrd-segment-replace-page (2 input fileGrps; overwrite segmentation below page of first fileGrp by all segments of second fileGrp, rebasing all coordinates; "inverse" of
replace-original
) - ocrd-segment-replace-text (insert text below page from single-segment text files; "inverse" of
extract-*
)
- comparing different layout segmentations:
- pattern-based segmentation (input file groups N=1, based on a PAGE template, e.g. from Aletheia, and some XSLT or Python to apply it to the input file group)
ocrd-segment-via-template
:construction: (unpublished)
- data-driven segmentation (input file groups N=1, based on a statistical model, e.g. Neural Network)
ocrd-segment-via-model
:construction: (unpublished)
For detailed behaviour, see --help
on each processor CLI.
For detailed description on input/output and parameters, see ocrd-tool.json or --dump-json
on each processor CLI.
Development
Prerequisities
Requires libgeos-dev
library for building shapely
binary requirement, see Shapely Installation from source. Please ensure it's available before trying to install local requirements.
Testing
None yet.